6. Artificial Intelligence Layer - eBattery AI

Traveling by electric vehicle requires making interconnected decisions. The distance to travel, the current battery level, the vehicle's consumption, road conditions, the location of charging stations and the driver's schedule are all parts of the same trip. When one of these variables changes, the other decisions may also be affected. A route that looks shorter can turn into a longer trip when suitable charging infrastructure is not available. A more powerful station may not deliver the expected time advantage if the charging power the vehicle can accept is limited.

eBattery AI sits at the center of the app to assess these decisions together and give the user an understandable trip plan. It lets the user express in natural language what they want to do; using the vehicle profile, map, route and accessible station data, it turns this request into actionable options.

The experience is simple for the user: saying “I want to go to İzmir”, “Plan a two-day trip for the weekend” or “Find a suitable charging station near me” is enough to get started. Behind the scenes, each request has to be matched with different data sources, calculations and app functions. eBattery AI's job is to manage this technical complexity while keeping the reasoning behind the decision clear to the user.

At the center of this approach is clarity, calculability and user control included. The assistant must be able to explain why the route it suggests is suitable; it must build range and charging calculations with a consistent method; and the user must be able to change the plan according to their own needs.

The current product’s AI chat and route and trip planning form the core of this structure. Inputs such as live station information, prices and traffic work together with the coverage of the relevant data sources. Forecasts enriched with weather conditions, direct vehicle connection and personalization that learns from behavior are capabilities that expand step by step on the same architecture.

6.1 Central Assistant Approach

6.1.1 Bringing the app's different functions together in a shared context

eBattery AI is a central assistant that reduces the need for users to give the same information again and again as they move between different parts of the app. A station search started on the map can be refined in the chat; a route created during the chat can be reviewed on the route screen; a planned trip can be edited, with its days and stops, in the trip module.

In this experience, the vehicle, trip and user preferences form the shared context. For example, when a user who has selected their saved vehicle says “Show me suitable stations nearby,” the system does not need to ask for the vehicle model again each time. The request is assessed together with the selected vehicle and the location information the user has chosen to share.

The centralized approach also ensures that different entry points in the app reach the same decision logic. A charging point search the user starts from the map and a search made through AI Chat should produce consistent results when the same vehicle, the same preferences and the same current data are used.

This way, eBattery’s modules are not just linked by a screen transition; the task the user wants to accomplish is also preserved.

6.1.2 Turning the user’s intent into an app action

People often do not express their needs in technical parameters. Sentences such as “I’m leaving within two hours,” “We’re going with the kids, so I’d like a more comfortable plan” or “I don’t want to detour too far for charging” contain several preferences at once.

eBattery AI turns these statements into information that can be used in planning. When needed, it fills in missing details such as the starting point, destination, battery level or time priority. Questions focus on the areas that genuinely affect the result to be produced.

For example, with the request "I want to go from İstanbul to İzmir", if the saved vehicle and starting location are already known, the user may be asked only for the current battery level and trip preference. When the user says "We can use toll roads, keep the total time short", this information is added to the route request.

This approach connects natural language to the app’s functions. Users can describe what they need without having to learn the technical structure of the menus; the system then carries the request into the relevant calculation and screen flow.

6.1.3 Completing the recommendation with user control

The plans offered by eBattery AI can be reviewed and changed. The user can change the recommended station, add an intermediate stop, remove a museum from the trip itinerary or ask for a lower-cost route.

The assistant relates these changes to the rest of the plan. Removing a stop does not only shorten the trip list; it can also change the daily distance, the arrival time and the charging need. The user's edit is evaluated while keeping these relationships intact.

The plan belongs to the user. AI makes options visible, simplifies calculations and explains the consequences of choices. The purpose of the trip, its pace and the acceptable cost are determined by the user.

6.2 AI Chat: Opening Experience, Avatar, Streaming, Context Memory

6.2.1 Opening experience: Showing what can be done right away

AI Chat's opening screen shows the user the assistant's core capabilities directly. The user can start a conversation with free text or use one of the ready-made starting options.

Starting option User’s goal Flow started
Plan Route Reaching a point by vehicle Creating a route based on start, destination, vehicle and charging needs
Plan Trip Exploring a region within a set time Preparing a daily schedule based on interests and time
Suggest Nearest Station Reaching a suitable charging point Generating recommendations from location, vehicle compatibility and accessible station data
My Efficiencies Understanding usage data Moving to consumption and usage assessment based on existing records

These options reduce uncertainty in the first interaction. Users do not have to guess what questions they can ask the assistant; the app offers starting points related to everyday EV needs.

As the conversation progresses, context-appropriate actions replace the starting options. After a generated route, actions such as “Open Route”, and after a suggested station, “Go to Station”, come to the fore.

6.2.2 Avatar: The assistant's visual identity

AI Chat's visual identity is an animated robot avatar. Before a chat begins, the avatar sits in the center of the screen and, with subtle movements and light effects, looks ready to interact.

When the first message is sent, the avatar shrinks and moves next to the assistant's messages. This shifts the screen's focus naturally from the starting experience to the conversation. More space opens up for what the user types, the assistant's replies and the result cards created.

The avatar also helps visually distinguish AI-generated messages. This identity creates an assistant experience that is distinct from, and consistent with, the community messaging in the app.

The function of motion and visual effects is to guide the user and make the state of the system understandable. Readability and interaction take priority; animations do not get in the way of essential information.

6.2.3 Streaming: Delivering the response as a stream

eBattery AI shows suitable answers as a stream. Instead of waiting for a long explanation to finish, the user can see the answer take shape step by step.

In this experience, two different processes are kept separate: writing the explanation and calculating the plan. The assistant starting to generate text does not mean the route calculation is complete. The route, station or trip card is presented when the relevant data and calculations are ready.

For example, the system may first show that the request has been understood, then share progress while preparing route options. When the calculation is complete, a map preview and charging stops are displayed. In this way, quick feedback is provided while the integrity of the results is preserved.

If the user changes the request during the flow, the new request takes precedence. A late-arriving result belonging to the earlier request should not replace the current plan. This is especially important in conversations where stops or route preferences are changed one after another.

6.2.4 Context Memory: Preserving the Meaning of the Conversation

Context memory makes it possible to understand the ongoing conversation together with its earlier steps.

When a user first says “I’m going from Istanbul to Izmir” and then writes “Plan the return too,” the assistant links the second request to the first trip. A phrase like “Make this more economical” is also applied to the route or trip plan being discussed.

Context is handled at three levels:

  • Chat context: Information provided and changes made during the conversation.

  • Active plan context: The route, trip or station selection being edited.

  • Saved user preferences: The selected vehicle and the settings the user has set permanently.

This distinction prevents a temporary preference from being interpreted as a permanent habit. A user saying "Let's avoid toll roads today" on one trip does not mean all of their future trips will be planned the same way.

Likewise, a battery level entered in a previous conversation cannot be used as current vehicle data the next day. Remembering context also requires managing how long the information remains valid.

6.3 Layered Architecture: Rule-Based Decision Layer and LLM Layer

6.3.1 Separating the roles of calculation and narration

eBattery AI’s architecture is built on solving different tasks with different methods.

Range calculation, energy needs, vehicle–station compatibility and the evaluation of route options are carried out in the decision layer, which works with defined data fields and calculation rules. Understanding the user's natural-language request and explaining the results are the job of the large language model, that is, the LLM layer.

This separation links the numerical basis of a recommendation with how it is explained. When the assistant says “This station is more suitable”, the reasoning should come from data produced by the calculation layer. The distance, time and energy values used in the explanation rely on the same source as the values on the result card.

Architectural component Core responsibility The result it produces
Data and integration layer Organizing data coming from vehicles, maps, stations and other sources Inputs with a defined source and freshness
Request and context management Combining the user's request, the selected vehicle and the active plan Structured planning request
Rule-based decision layer Running compatibility checks, energy calculations and candidate evaluation Route and charging plan options
LLM layer Interpreting natural language and explaining calculated results Explanation suited to the user
Verification and action layer Checking the consistency of results and transaction authorization A displayable card and an actionable step
Experience layer Presenting the chat, cards and screen transitions In-app user experience

6.3.2 Scope of the rule-based decision layer

The rule-based approach means that, for the same data snapshot and the same rule version, calculations can be reproduced consistently.

This layer calculates the usable battery energy, estimates the energy needed for route segments, eliminates stations that are incompatible with the vehicle, and compares suitable candidates using defined criteria. The road preferences the user selected and the battery reserve to be kept are also taken into account here.

Rule-based calculation does not mean the outside world is known with certainty. Future traffic, actual consumption or the station’s condition at the time of arrival may change. Therefore, even if the calculation method is consistent, some inputs and outputs are estimates.

eBattery AI preserves this difference: known data, information entered by the user and a model estimate are not presented with the same level of certainty.

6.3.3 Scope of the LLM layer

The language model is used to understand the user's request and to turn the calculated result into a natural explanation.

Phrases such as “I want less walking”, “Remove the museums”, “I don't want to wait too long for charging” or “Why did you choose this station?” fall within this layer's working area.

The LLM helps produce a structured request that can be passed to the appropriate app function. However, it does not fill in the station’s price, socket status, or the vehicle’s battery level from its own general knowledge. This information is taken from the relevant records and integrations.

For example, when the decision engine selects a station because it involves less route deviation and suitable power support, the assistant can put this reasoning in simpler terms:

“This station is compatible with your vehicle and closer to the main route. The alternative that looks more powerful does not provide an extra time advantage because of your vehicle’s charging limit.”

The value of this explanation is that the user can see the logic behind the decision.

6.3.4 From verified data to an explainable result

Each result is linked to the key inputs it used. The selected vehicle, battery information, calculation time and the freshness of station data make it possible to understand the conditions under which the plan was produced.

This structure also makes it easier to evaluate user feedback. A recommended station not being chosen does not always mean the calculation was wrong. The user may want a different break point, may prefer a certain operator, or may have a time constraint that does not appear in their plan.

Examining the inputs of the decision together with the user’s change helps the system be improved in a more meaningful way over time.

6.4 Route Planning — Vehicle Profile, Range Calculation, Route and Road Preferences

6.4.1 Planning the trip together with the vehicle profile

In eBattery, route planning takes the vehicle's energy and charging characteristics into account in addition to the start and end points.

Two electric vehicles on the same route may need different charging plans. Usable battery capacity, consumption, supported sockets and charge acceptance power create this difference.

The main vehicle details used in planning are as follows:

Vehicle information Role in planning
Brand, model and suitable vehicle variant Ensures the correct technical profile is selected
Usable battery capacity Forms the basis of the energy calculation that can be used on the trip
Consumption value Converts distance into an estimated energy need
AC charging support Helps evaluate long-break and destination charging options
DC charging support Makes it possible to evaluate en-route fast charging options
Socket compatibility Enables identification of available stations
Current battery level Defines the trip’s starting energy state

Gross battery capacity and usable capacity are assessed separately. The value used as the basis for range calculation is the energy available for driving.

In the current usage flow, the user enters the battery level manually before the journey. While the saved vehicle profile carries the technical information, the battery percentage is the current input for that journey. Future vehicle-connection integrations form the room for expansion that would let this information be taken automatically from supported vehicles.

6.4.2 Range calculation and energy reserve

The range calculation is based on the relationship between available energy and expected consumption:

Current energy = Usable battery capacity × Battery level / 100

Base range estimate = Available energy / Consumption × 100

Here, energy is in kWh, consumption in kWh/100 km and range in kilometers.

As an example, consider a vehicle with a usable battery capacity of 60 kWh, a current charge level of 40% and a consumption of 18 kWh/100 km used in the calculation.

In this case:

  • The calculated energy in the battery is 24 kWh.

  • The theoretical range down to zero is about 133 kilometers.

  • For example, if a 15% arrival reserve is kept, the energy available for planning drops to 15 kWh.

  • With the same consumption assumption, the distance calculated while keeping the reserve is approximately 83 kilometers.

This example explains why the theoretical range shown on screen differs from the distance that can be used for planning. The 15% here is an example reserve used to show the calculation logic.

In the actual route calculation, distance is assessed together with road characteristics and the available consumption inputs. As data coverage expands, effects such as gradient, traffic and weather conditions may be added to the model. When the conditions the estimate relies on change, the plan is re-evaluated.

6.4.3 Start, destination, and intermediate stops

The user can set the starting point from their current location, an address search or the map. The destination and intermediate stops are also parts of the same trip plan.

Intermediate stops are especially important in EV planning. When traveling from İstanbul to İzmir, a visit in Bursa changes more than just the total distance. The duration of the visit, whether charging is available in that area, and the battery level at which the next leg begins also affect the plan.

The user can change the order of the stops or keep a particular order. A manually set order can represent the user’s schedule or personal priority. The system recalculates the charging stops in line with this preference.

With each change, the relevant plan is updated without the user having to re-enter all the information.

6.4.4 Route mode and road preferences

eBattery links route options to the user’s trip priority.

Fastest approach focuses on total trip time. In addition to driving time, detours made for charging, charging time and, where data is available, waiting time also enter the assessment.

Most Efficient approach addresses the energy need and the available cost data together with a reasonable trip duration. While offering the user a lower-cost option, its effect on duration is also shown.

Neither of these two options has a fixed advantage for every trip. On a business trip, time may matter more; on a weekend trip, the break schedule or budget may.

Route preferences such as toll roads, highways and ferries are added to the route request. The Google Maps Routes API also offers avoidance options for such preferences; these options steer the route search, and whether a suitable alternative can be found depends on the road network. eBattery combines these capabilities, provided by the map service in use, with vehicle and charging planning.

Regional or more detailed road restrictions are handled together with the coverage of the relevant data and routing service. The route presented to the user should clearly show the effect of preferences on the result.

6.5 Charge Planning and Station Scoring — Charge Amount Recommendation, Safety Margin

6.5.1 Compatibility first, then comparison

The first step of charging planning is to identify the stations that can be used. A station having a high rating or a low price does not mean the vehicle can charge there.

For this reason, the assessment is carried out in two stages:

  1. Identifying candidates compatible with the vehicle and the trip.

  2. Comparing suitable candidates in terms of time, cost and usage conditions.

Socket compatibility, accessibility and the energy needed to reach the station are the subject of the first stage. These basic conditions are not treated as scoring factors that can be offset by other advantages.

For example, a station that is incompatible with the vehicle is not recommended just because it is cheaper. Similarly, a point that does not seem reachable with the current energy does not gain priority solely because of its high charging power.

6.5.2 Station scoring: Suitability that varies by trip

Once suitable candidates are identified, stations go through a multi-criteria evaluation.

Criterion Topic assessed
Route deviation The extra distance and time needed to reach the station and return to the main route
Power compatibility with the vehicle How the station's power relates to the range the vehicle can use
Price The tariff that is accessible and applicable to the relevant user
Current status Current status of a suitable charging point in supported integrations
Waiting information Likely wait time when reliable data or sufficient model support is available
Operational reliability Service continuity and usage experience when data is available
User reviews How up to date experiences are, what they contain, and whether they are based on a sufficient number of observations
Break fit The station’s relationship to the user’s current stops and time plan
Data freshness How up to date the information used in the decision is

The importance of these criteria varies with the selected route mode. On a time-priority trip, detours and charging time may be more decisive. On a cost-priority trip, a tariff difference is assessed together with the extra time it would cause.

Missing data is also part of the assessment. A station whose live status is unknown is not treated as "available." Likewise, socket occupancy information alone is not always enough to calculate a reliable queue time.

A user rating alone is also not a substitute for operational reliability. A few old reviews should not be given the same weight as a current and broader set of experiences.

6.5.3 The relationship between station power and the actual charging experience

The power value shown on a station does not mean the vehicle will charge at that power for the whole session. Charging time depends on multiple factors, such as battery level, capacity, and the characteristics of the vehicle and charging equipment. The U.S. Department of Energy's Alternative Fuels Data Center resource likewise explains that these variables together determine charging time.

For example, for a vehicle with a DC charging acceptance limit of 100 kW, a station with a capacity of 350 kW may not be faster than a suitable station with a capacity of 150 kW, simply because of its nameplate power. If reaching the first station requires a longer detour, the second option may be more suitable for the journey as a whole.

eBattery's assessment considers station power together with the vehicle's characteristics. Instead of presenting high numbers automatically as the better option, it looks at the benefit the station would bring to the user's journey.

6.5.4 Charge amount: Energy suited to the next leg

The charging plan covers the recommended amount of energy as well as the choice of station.

The basic approach is to consider together the energy needed for the next route segment, the targeted arrival reserve and suitable alternatives. The user does not need to reach the same battery level at every stop. A short hop, a long next leg or charging availability at the destination can create different goals.

The basic relationship for the energy that needs to be added to the battery is as follows:

Energy to be added = Usable capacity × (Target battery percentage − Current battery percentage) / 100

For example, in a vehicle with 60 kWh of usable capacity, raising the battery level from 20% to 70% corresponds to an energy increase of approximately 30 kWh in the battery.

This value does not have to be exactly the same as the energy measured at the station meter or billed. Because of charging losses and auxiliary consumption, the point at which the energy is measured matters.

Charging time is also not calculated simply by dividing by the station's maximum power. Where accessible, the model takes into account the vehicle's charging behavior at different battery levels. As an illustrative example, if 30 kWh of energy is added to the battery at an average net power of 60 kW, the energy transfer time is about 30 minutes. The deciding value here is the average power during the session.

In the cost calculation, applicable session, time or other tariff components are taken into account in addition to the energy fee. The user should be able to understand what scope the displayed estimate covers.

6.5.5 Preserving the safety margin and alternatives

Planning is not based on using the battery theoretically down to the last unit of energy. It leaves room for an arrival reserve, variability in consumption estimates and access to alternatives along the road.

The required reserve is assessed together with route length, station density, data freshness and the flexibility of the plan. An urban route with frequent charging options and a long-distance section with limited alternatives do not carry the same conditions.

When eBattery AI explains the recommended charge amount, it makes this logic visible. The user can understand why taking more energy at a station is recommended, or why a short charge is considered enough to reach the next suitable stop.

Current availability information also reflects the situation at the time the plan was created. This information does not create a reservation that the same socket will be free on arrival. Evaluating alternative stations is therefore an important part of the plan.

6.6 Dynamic Recalculation

6.6.1 Reflecting changing conditions during the trip in the plan

The route is not treated as a fixed document created before the trip begins. As the user progresses, location, traffic, station status and their own preferences may change.

Dynamic recalculation assesses the effect of these changes on the plan. However, each trigger works only to the extent that the relevant data is accessible.

Change Information used Possible impact on the plan
Change in traffic Map and traffic source Updating duration and route options
Change in the status of a planned station Operator integration Evaluating an alternative charging point
The user deviating from the route Location and route progress Recalculating the remaining distance and energy need
Updating battery information User sign-in; supported future vehicle connections Updating the charging stop and the target energy amount
Adding or removing an intermediate stop User editing Re-planning the affected route sections
Changing weather conditions The scope in which the relevant integration is active Re-evaluating the consumption estimate

On a trip that runs on a manually entered battery level, the remaining energy calculated by the system is an estimate. It is not labeled the same way as current battery data measured from the vehicle.

6.6.2 Not presenting every change to the user as a new route

Small differences that keep changing can place an unnecessary decision burden on the user. A few seconds of traffic fluctuation or a very small time advantage does not require a new recommendation every time.

The recalculation approach assesses what the change means for the journey. The feasibility of the current plan, the energy reserve and the clear benefit to the user take priority.

This keeps the plan stable. Instead of being steered constantly between two very close options, the user sees a meaningful update when one is needed.

6.6.3 Explaining the reason for the change

The suggested new route is presented in comparison with the current plan. The user should be able to see which condition has changed and what the result is.

For example:

“Usage information has changed for the suitable connectors at the station you planned. An alternative station is closer to your route and reachable based on the current energy estimate. You can review the new option.”

Only elements supported by the available data are used in the explanation. If there is no reliable wait-time estimate, no exact minute saving is produced.

Traffic information from the map provider and eBattery's EV assessment based on this data are also linked correctly. The reasoning behind the recommendation makes clear which data is used and how.

6.7 AI Trip Planning — Candidate Pool, Clustering, Time Budgeting, Charging Integration

6.7.1 From transport plan to everyday experience

Trip planning organizes the time the user will spend in a given area. The goal is to turn places of interest into a schedule that fits the user’s time, interests and the energy needs of their vehicle.

When the user says “I’ll spend two days in Cappadocia, focused on nature and history,” the system evaluates the region, the trip length and the priorities together. The starting point, the daily pace and, if there are any, the places the user specifically wants to see are added to the plan.

The resulting schedule provides an editable trip structure divided into days.

6.7.2 Building the candidate place pool

The first step is identifying suitable places to visit in the region. The candidate pool may include cultural sites, museums, natural areas, food and drink spots and rest stops.

Depending on the scope of the source, the information taken into account includes location, category, visit duration, opening hours and user reviews.

High popularity alone is not a reason for selection. If a place does not match the user’s interests, is too far from the daily schedule, or there is not enough time to visit, it may receive a lower priority.

Content narration is tied to the records of the selected places. Information that can change, such as opening hours or entry conditions, is not generated from the language model's general knowledge as if it were current data.

6.7.3 Geographic clustering and transport relationships

Candidate places are assessed for their geographic and transport relationships before the daily schedule is built. The aim is to reduce unnecessary back-and-forth trips within the same day.

Two points that look close together on the map may not be a short trip apart. Road connections, crossing points and the geography of the area determine the real movement pattern. For this reason, daily grouping is not based only on straight-line distance.

For example, while a day's schedule is built from visits in the same area, a farther stop can be moved to another day. This way, the user can spend more of their time on visits rather than travel.

When the user prefers a certain order, this choice is preserved; the system adapts the remaining plan to be consistent with that order.

6.7.4 Time budgeting: A realistic daily schedule

Whether the daily plan is workable depends on travel and break times as well as the total visit duration.

A day's time budget includes the following components:

  • Travel between points.

  • The visit time spent at each place.

  • Meal and rest breaks.

  • Required charging stops.

  • A reasonable time allowance for parking, access and similar transitions.

Placing only stops whose visit times add up to six hours into a user's six-hour trip plan leaves no time for travel. eBattery AI builds the plan within the day's total time frame.

The program's value is not measured by listing as many places as possible. What matters is that it can be followed at the pace the user chooses and leaves room for changes.

6.7.5 Fitting charging into the daily schedule

Charging stops are considered together with the energy need when a trip plan is created. They are not treated as a separate need added at the end of the day.

If a meal break can be matched with a suitable charging point, the two activities can take place in the same time window. For example, a 40-minute meal break and a 25-minute charge that start at the same place and time do not, under suitable conditions, require a total 65-minute additional stop. The plan evaluates overlapping durations together.

In this matching, the station's actual access, operating conditions, parking layout and the user's preferences matter. A station that appears close to a restaurant on the map is not assumed to be a suitable break option in every case.

For trips spanning multiple days, the energy calculation carries on across days. The next day's starting battery is linked to the previous day's planned end. Overnight charging is factored in only when it is included in the user's plan and supported by the relevant facility.

6.7.6 Editable trip narrative

The calculated schedule is presented as stops divided into days, along with an explanation in natural language. The user can see which area will be visited on which day, where charging is placed, and the approximate time breakdown of the day.

Requests such as “Make the second day more relaxed,” “Remove the museum,” or “I want to be in nature in the afternoon” let you edit the plan.

After a change, the affected days and route segments are reassessed. The trip narrative also changes with the updated plan; the principle is to avoid any contradiction between the old description and the new list of stops.

This city- and region-focused structure forms the basis for broader, country-scale and long-term planning in later stages.

6.8 Efficiency Analysis

6.8.1 Making data meaningful from the driver’s perspective

Efficiency analysis helps users assess their energy and trip records more clearly.

If the accessible data is sufficient, consumption, distance, charging spend and planned versus actual values from different periods can be compared. AI explains the change within the numbers and shows which records it may be useful for the user to review.

The scope of the analysis depends on the data used. Information obtained through a vehicle connection, session records or user sign-in is kept separate from estimates calculated only from the route.

6.8.2 Measured consumption and calculated estimate

In a consumption assessment, the energy and distance must belong to the same period, the same vehicle and a consistent measurement method.

If total energy and total distance are known, the basic consumption calculation is as follows:

Average consumption = Total driving energy / Total distance × 100

Simply adding up the consumption values of trips of different lengths and dividing by the number of trips may not give the correct average for the period. Calculating from total energy and distance preserves the weight of short and long trips.

Energy purchased during charging is also not always the same as the energy consumed while driving in the same period. Battery levels at the start and end of the period, charging losses and the scope of records affect this relationship.

For this reason, eBattery treats charging spend, purchased energy and driving consumption as separate concepts.

6.8.3 Showing the change and explaining its cause with data

As an illustrative example, suppose that average consumption rises from 18 kWh/100 km to 19.8 kWh/100 km across two comparable periods. This change means a 10% increase in consumption.

However, this result alone does not show the cause of the increase. The route, speed, weather conditions, load or trip structure may have changed.

The AI's assessment stays at the level the data supports. If only consumption and distance records exist, it can say "Your consumption has increased"; reaching the conclusion that a specific behavior such as hard braking or air conditioning use caused it requires the relevant measurements.

Where there is no direct vehicle data, battery health is not presented as if it had been measured. Detailed driving behavior analysis expands with integrations that provide enough data to evaluate these behaviors.

6.8.4 Building comparable usage habits

The purpose of the efficiency screen is to make it easier for users to understand their own usage over time.

Comparing similar routes, reviewing charging spend at different stations, and seeing the differences between planned and actual duration support this understanding.

The assistant’s suggestions are concrete and measured. Instead of promising the user an unverified savings percentage, it explains which data has changed and which conditions should be examined for a more meaningful comparison.

6.9 Result Cards and Deep Linking

6.9.1 Turning a conversation into a result that can be used in the app

A route, trip or station suggestion created in AI Chat is presented with result cards. The cards let the user see the key information quickly and move to the relevant module.

Card type Card’s core content Related screen
Route card Start, destination, route summary, estimated time and charging stops Route Planning
Trip card Region, number of days and daily schedule summary Trip Planning
Station card Location, compatibility, power and available current information Station Details
Efficiency card Period, key indicators and assessment summary Efficiency

The purpose of the card is not to squeeze all the details into a small space. It offers enough information for the user to recognize the result and choose the next action.

6.9.2 Opening the same plan with deep linking

Deep linking makes the result card take the user to the right content on the relevant screen.

When the user opens a route they created in the chat, they do not encounter an empty route form. The start and destination points they selected, the vehicle, intermediate stops and the plan's other details appear on the relevant screen.

The same relationship is preserved on trip and station cards as well. This way, the chat, map and planning screens become different views of the same action.

Technically, the card is linked to the relevant record and plan version. When the user changes the plan, it stays clear which result is current.

6.9.3 Context-appropriate quick actions

Depending on the results, actions such as “Start Route”, “Go to Station”, “Save”, “Share” or “Add to Favorites” are offered.

Not every card needs to show all actions. Actions are chosen according to the result the user has produced at that moment.

Shareable cards are also handled together with their content scope. Information such as the user's home address or personal starting point is not considered a natural or mandatory part of sharing. The user should be able to see what content they are sharing.

6.10 Proactive Suggestions and Notifications

6.10.1 Providing meaningful support when a need arises

When the relevant data is available, eBattery AI can also surface important changes without the user asking.

Within this scope are a change in station status during an active trip, a significant traffic impact, or re-evaluating the charging plan based on updated battery information.

The purpose of proactive support is to help the user in a timely way. Not every data change creates a notification; how the information relates to the current trip and whether it requires action is assessed.

6.10.2 Evaluating a Notification Together with the Data It Is Based On

The meaning of a low battery warning depends on the source of the data.

If the current battery level can be obtained from the vehicle, the measured value can be used. If the user entered it manually at the start, the value calculated during the trip is an estimate. The wording of the notification preserves this difference.

Similarly, to generate a suggestion about the price advantage of a favorite station, the current tariff and the conditions applicable to the user must be known. A maintenance reminder, on the other hand, is tied to a date saved by the user or to a suitable data source.

This approach keeps proactive recommendations consistent with the available data scope.

6.10.3 Notification frequency and user preferences

Notifications are managed according to their importance and context. A change that affects the feasibility of an active trip does not carry the same priority as a general usage suggestion.

Users can manage their notification preferences. Reducing repeated messages for the same event, not showing notifications that are no longer valid, and directing the recommendation to the right plan are part of the experience.

Location and background functions run with the user’s permissions and within the operating conditions of the mobile operating system. Information shown during a trip is kept short and action-oriented; details can be reviewed on the relevant screen.

6.11 Trust and Error Handling

6.11.1 Making the nature of the information visible

eBattery AI’s approach to trust is built on making it possible to understand what information the results are based on.

The battery level entered by the user, the socket status received from the operator, the map service’s time estimate, and the AI’s explanation are different types of information. This distinction allows the recommendation to be evaluated correctly.

Generative AI systems can present unsupported information in a convincing way. NIST's generative AI risk profile also addresses this behavior. eBattery's data-based calculation and verification approach is designed to reduce the impact of such errors on app decisions.

This approach matters especially for location, station status, price and energy calculation. A fluent explanation is not used in place of verified data.

6.11.2 Behaving clearly and consistently when data is missing

If one data source does not respond, it does not mean all of the app's functions must stop at once. The system assesses what result can be offered with the data available.

For example, if a station’s location and socket information are available but its live status cannot be retrieved, station discovery can continue, while availability is shown as unknown.

Similarly, when the language model is temporarily unable to respond, the valid route result received from the calculation service is parsed so that it can be presented with a standard card and a short explanation.

Where the last known data is used, freshness is preserved. An old status shown from cache is not presented as live information that has just been received.

6.11.3 Authorization, privacy and external content security

The data the AI layer needs is limited to the scope of the task to be performed. Explaining why a station is recommended does not require passing the user's entire message history or payment information to the model.

Location, travel plans and user preferences are handled under access control. The app's transaction permissions are verified independently of the language model's free-text generation.

Descriptions coming from external sources and user reviews are also processed as data. One of the security principles of the architecture is not allowing instructions that may appear in this content to change the application's transaction rules.

In future transaction flows such as payment or reservation as well, an AI suggestion and the execution of the transaction are separate steps. The scope of the transaction is shown to the user and the relevant approval flow is preserved.

6.11.4 Learning from errors in a measurable way

Reliability requires regular assessment of real usage scenarios.

Situations such as a station incompatible with the vehicle, a missing price, a changed battery input, a connection interruption or an old route request that completes late are test scenarios that determine how the system should behave.

User feedback also contributes to this process. A choice made on wrong station information and a choice that did not meet expectations are not the same type of error. Matching feedback to the relevant data and decision step helps improve the right component.

The trust approach goes beyond counting results only as "successful" or "failed"; it is based on understanding the source of the error and its effect on the user.

6.12 Model-Independent Infrastructure and Performance Targets

6.12.1 Keeping product logic independent of the model provider

The infrastructure of eBattery AI is based on an approach that does not tie core product functions to the features of a single language model.

Vehicle profile, station data, route calculation and user preferences are managed in the app's own services. Language models connect to these services through defined tasks and data contracts.

This structure allows different model families such as GPT and Gemini, or alternatives developed in the future, to be evaluated. A change of model should not require rewriting core rules such as vehicle–station compatibility.

Models are selected on language quality, structured output accuracy, task success, latency, cost and data processing conditions. The provider's name alone is not used as a measure of product quality.

6.12.2 Using the right model and service for the task

Not every user request requires the same processing load. A short intent classification and a detailed trip narrative have different needs.

The infrastructure is designed so that tasks can be routed according to this difference. Work that calculation services can solve directly is carried out in those services; an appropriate model is used for steps that require natural-language interpretation and explanation.

The architectural approach that supports performance includes the following elements:

  • Running independent data calls concurrently where appropriate.

  • Controlled caching of vehicle information that changes more slowly.

  • Applying freshness rules to variable data such as availability and price.

  • Canceling invalid accounts when the user’s request changes.

  • Running long operations in the background without blocking the interface.

  • Presenting the ready parts of the result in stages, while preserving its integrity.

Map and traffic services also involve trade-offs between response time and calculation detail. The Google Routes API defines different options for traffic assessment and explains that a more comprehensive search can create higher latency. eBattery's performance approach evaluates the behavior of such external services within the overall user experience.

6.12.3 Defining performance targets

In eBattery AI, performance is assessed through the experience between the user submitting a request and reaching a usable result.

The start of the first response and the completion of the route plan are different measurements. Showing a loading animation is not treated as the same thing as the start of a meaningful AI response.

Performance area Design objective Measurement approach
AI response starting Start the first meaningful response stream in under 2 seconds under standard conditions The time from sending the request to the first meaningful content visible to the user
Standard route creation Delivering a usable route result in under 5 seconds wherever possible in typical supported scenarios Time from the request until the route and basic charging plan are ready
Screen transitions Keeping interaction and fluidity while a calculation is running Launch delays and interface freezes
Recalculation Reflecting meaningful changes in the plan in a timely manner The time from receiving the update until the new option is ready
Complex trip plans Delivering results in stages by keeping progress visible First usable result and completed plan times

These values are engineering targets. Actual results are measured according to network conditions, data providers, route length, number of stops and the complexity of the plan.

A single average does not describe the whole experience. In addition to the typical user experience, measurements such as the median and the 95th percentile are tracked so that slower requests can be seen. This makes visible the gap between the system running fast for most users and some users waiting a long time.

6.12.4 Measuring accuracy and usability together with speed

Performance evaluation covers the quality of the result as well as the response time.

The main areas of assessment are:

  • Accuracy of selecting stations compatible with the vehicle.

  • Consistency of energy and battery percentage calculations.

  • Compliance with the stop order set by the user.

  • Keeping current data and old data distinct.

  • The description text matching the values on the result card.

  • The result card opening the correct plan and screen.

  • Applying user changes to the correct version.

  • Consistent understanding of the same request across different languages.

When actual data becomes available, the estimated arrival time is compared with observed results for consumption and charging time. These comparisons show in which regions, vehicle profiles or usage conditions improvement is needed.

Model updates are handled with the same evaluation discipline. A new model speaking more fluently does not mean it is more accurate in route explanations. Its performance on current tasks, its structured outputs and its consistency with in-app actions are reviewed together.

6.12.5 A common foundation for expanding capabilities

The model-independent structure supports eBattery AI expanding with new data and services as the product develops.

Direct vehicle data, including weather conditions in the energy model, recommendations that learn from user preferences, and trip plans with wider geographic coverage are all tied to the same calculation and context structure.

As charging-payment, EV Market, mobile charging support and other services come online, the actions the assistant can guide also expand. Each new function is handled together with the real scope, data and user approval of the relevant service.

Throughout this development, the relationship the assistant builds with the user is preserved: understanding the need, using the appropriate data, explaining the consequences of the options, and making the action the user selects easier within the app.

Search eBattery

Search for a feature, guide or topic.

Starting pointsFinding a charging station →Sample RWA project →Guides and news →