Built for the All Things Agentic Hackathon
Electric vehicle adoption is growing rapidly, but the charging experience is still fragmented. A driver may find a charger on a map, yet still has to think about battery percentage, connector compatibility, charging speed, queue time, cost, and what happens if the charger they were depending on suddenly becomes unavailable.
That problem became the starting point for Vidyut.
Vidyut is an agent-driven EV charging and journey-management prototype that connects EV owners, property hosts, charging-network operators, and administrators through separate role-scoped workspaces.
The goal is not simply to show chargers on a map. The goal is to make the charging ecosystem capable of planning, monitoring, reasoning, and safely reacting when real-world conditions change.
A Multi-Role EV Charging Ecosystem
Vidyut begins with one common platform but provides a different experience depending on who is using it.
EV owners manage vehicles and journeys. Property hosts can make their locations available for charging infrastructure. Charging companies manage their networks and faults. Administrators get an isolated control plane for governance and monitoring.
Vidyut — an intelligent EV charging ecosystem connecting EV owners, hosts and charging providers.
Instead of treating these participants as unrelated systems, Vidyut brings them into one coordinated ecosystem.
The EV Owner Experience
An EV journey depends heavily on the vehicle being used.
Different EVs have different battery capacities, energy efficiency, charging speeds and connector compatibility. Vidyut therefore maintains an Autonomous Garage where users can manage multiple vehicles.
Multiple EVs can be managed with individual battery and connector characteristics.
For the prototype, vehicle telemetry is clearly marked as demo telemetry, allowing battery-aware flows to be demonstrated without claiming a connection to physical vehicle hardware.
Demonstration telemetry provides current battery percentage, estimated range and charging state.
This battery state becomes an important input when Vidyut evaluates whether a charger is actually reachable.
Charging Is More Than Finding a Station
A charging platform also needs to coordinate what happens after a station is selected.
Vidyut includes booking, active-charging, wallet and vehicle-linked auto-recharge experiences.
Confirmed reservations can transition into charging sessions using compatible available connectors.
Vehicle-linked wallet and auto-recharge rules are modeled independently for each EV.
These features are useful, but the most interesting part of Vidyut appears when something goes wrong.
What Happens If a Planned Charger Fails?
Imagine an EV already has a charging-aware journey planned.
The vehicle is travelling toward a specific connector.
Then that connector goes into a FAULT state.
A conventional application might simply display:
“Charger unavailable.”
Vidyut instead starts a recovery workflow.
The system evaluates the failed connector, current vehicle state, remaining route and battery safety constraints, then searches for an alternative that can actually be reached.
CHARGER FAILED — REMOVED FROM JOURNEY
When the exact reserved connector fails, Vidyut removes it from the active journey and constructs a replacement route.
This distinction matters.
A charger should not be considered safe merely because it is geographically nearby. The vehicle must be able to reach it while respecting its configured minimum battery reserve.
Self-Healing Journey Recovery
In the demonstrated recovery, Vidyut selects a replacement charging connector and recalculates the journey.
The updated plan exposes:
- battery level at evaluation,
- predicted battery at the recovery charger,
- minimum reserve,
- charging target,
- next charging stop,
- remaining distance,
- cost difference,
- ETA impact.
Vidyut recalculates the remaining journey after the replacement charger is approved.
The screenshot demonstrates a particularly important architectural principle:
AI can reason about the options, but deterministic backend checks remain responsible for safety constraints.
The agent should never be able to simply invent an unreachable charging stop.
The backend first establishes what is safe and valid. The AI agent then reasons over those acceptable choices.
Human Control Over Agent Actions
Vidyut also models different levels of autonomy.
For example, the Company workspace exposes an Ask Before Actions mode.
The AI can investigate a fault and prepare an operational response, but consequential changes still require approval.
Vidyut separates AI reasoning from permission to execute state-changing actions.
This approach is useful because autonomous systems do not need to be all-or-nothing.
An agent can monitor continuously, analyze automatically and prepare an action while still requiring a human for the final decision.
The Property Host Agent
EV infrastructure also depends on finding places where chargers can actually be installed.
Vidyut therefore gives property owners a dedicated Host workspace.
The Host workspace combines property, charger, booking and network-health information.
Hosts can maintain installation-ready properties and expose them to charging operators.
The property model can include factors such as:
- parking bays,
- available electrical load,
- power type,
- operating hours,
- location,
- verification status.
The Host Agent can then help compare those locations.
The Host Agent evaluates property-readiness information and explains its recommendation.
In the example shown, the agent analyzes the available properties and ranks them using backend data such as parking capacity, electrical load, grid supply and operating hours.
The interface also visibly identifies the response as coming from Gemini 3.6 Flash.
Connecting Property Owners and Charging Operators
Property owners may have strategically valuable locations but no charging infrastructure.
Charging companies may want to expand their network but need suitable locations.
Vidyut attempts to connect those two sides through a property marketplace.
Charging operators can inspect potential low-density charging areas and verified host sites.
This creates a path from:
Property availability → site evaluation → operator discovery → charger deployment → customer charging
instead of treating infrastructure expansion as a completely separate process.
The Charging Company Agent
Charging operators require a different form of intelligence.
Their priorities include connector health, faults, utilization, bookings and maintenance.
The Company workspace surfaces connector-level operational faults across the operator's network.
The Company Agent can inspect the affected infrastructure and prepare actions such as maintenance or booking restrictions.
The Company Agent evaluates operational faults using only data available inside the authorized company workspace.
Again, the AI operates inside a defined scope rather than receiving unrestricted access to the entire platform.
An Isolated Administrative Control Plane
Some actions require stronger governance than an ordinary user workspace.
Vidyut therefore includes a separate Admin portal.
Administrative access is isolated from EV Owner, Host and Company accounts.
This layer is intended for functions such as verification, support, operational governance and auditing.
Running the Agentic Backend on Google Cloud
The application is not only a local prototype.
The backend services used for the demonstration are deployed on Google Cloud Run.
The Vidyut AI agent running as a Google Cloud Run service.
The agent layer is built using Google Agent Development Kit (ADK) with Gemini 3.6 Flash as the primary reasoning model.
Vidyut's application backend deployed independently on Google Cloud Run.
Separating the transactional backend from the agent service allows the system to maintain an important boundary:
The agent reasons.
The backend validates.
The user controls consequential execution.
What I Learned Building Vidyut
One of the biggest lessons from the project was that making an application agentic should not mean handing every decision to an LLM.
For EV routing in particular, certain constraints should remain deterministic.
Battery reserve, road reachability, connector availability, authorization and transaction validity should be verified by application logic.
The AI is valuable for understanding intent, comparing alternatives, explaining decisions and coordinating tools.
Combining those two approaches produces a much more dependable architecture than relying exclusively on either one.
Another important lesson was that infrastructure failures should be modeled as part of the journey itself.
A route is not really intelligent if it only works while every charger behaves exactly as expected.
From Charger Finder to Agentic EV Operations
Vidyut started from a simple question:
What should an EV application do when the charger a driver is depending on suddenly fails?
That question expanded into a larger system connecting:
EV Owner Agent → Host Agent → Charging Company Agent → Governance Layer
with charging-aware routing and recovery at the center.
The result is a hackathon prototype that explores what EV charging could look like when software does more than display infrastructure.
It can understand a journey, coordinate charging, observe infrastructure changes, reason about alternatives and safely recover when the original plan fails.
That is the direction behind:
Vidyut — Powering a Smarter Tomorrow
Prototype Disclaimer
Vidyut is a hackathon prototype. Charging stations, tariffs, commercial relationships, faults, vehicle telemetry and marketplace activity shown in the demonstration include synthetic demo data and should not be interpreted as real-world deployments or affiliations.
Built for the All Things Agentic Hackathon using Google ADK, Gemini and Google Cloud.
Top comments (0)