AI that runs inside the vehicle, not just in the cloud behind it
A voice copilot that keeps answering when the network drops, and hours of drive footage turned into findable, speaker-labelled moments — both built on real vehicle hardware, offline-first by design.
The problem
In-vehicle AI that assumes constant connectivity fails exactly where it matters most, on Indian roads with patchy coverage. And the record of the drive is no better: hours of footage are unsearchable, so a specific incident or conversation can only be found by watching the whole clip back.
What the portfolio proves
On-device and offline-first AI on real vehicle hardware and real drive data: local model inference with a cloud fallback, voice interaction, speech transcription and long-form video understanding that survives a dropped connection.
How the projects fit together
xPilot (Prototype) is a tap-to-talk Android copilot that answers over a live map using on-device rules and a local model first, falling back to a cloud model only when needed. DashCam (In use) ranks, transcribes and speaker-labels hours of dashcam footage and ships an on-device Android companion that does the same recall entirely on the phone. The AutoMotivate workshop (Prototype) — the build behind AutoMotivate, EmbedAI's AI upskilling practice for automotive R&D — sits here as secondary domain evidence: the same automotive engineering grounding, applied to upskilling rather than to a shipped app.
What I can do here
Where a client wants AI running at the edge rather than only in a cloud dashboard, this is the evidence the Services conversation points to.
How mature this is
Neither project is a shipped consumer app. DashCam's own policy blocks publishing any footage, so this vertical is written and illustrated in text and abstract visuals only — no frame, still or clip is shown or implied.
Mapped projects
AutoMotivate workshop
Prototype
The core works end to end and can be demonstrated; it is not production-hardened or generally available.
A delivery-ready two-day GenAI workshop build for automotive R&D teams, and the one carded build behind AutoMotivate, EmbedAI's AI upskilling practice for automotive R&D.
Automotive engineering teams are told to adopt AI, then handed a generic course built for office workflows, not their own kind of work.
An 11-module, 22-live-demo executive workshop with a full delivery kit (run-of-show, preflight checklist, governance pack) covering GenAI-assisted coding, ECU log analysis, agentic workflows and PII-safe automation, all on synthetic automotive data for a fictional company.
Built on LangGraph and CrewAI for the agentic-workflow modules, with a PowerShell workshop runner orchestrating the demo scripts.
15 or more synthetic automotive artefacts are all authored for a fictional company, so the demo content itself is safe to describe in full.
PowerShell workshop runner
Python demo scripts
LangGraph
CrewAI
DashCam
In use
Running today and in regular use; it is not an open public service you can sign up for.
A pipeline that ranks, transcribes, speaker-labels and mood-tags dashcam footage, so a moment can be found by the words spoken. A companion Android app does the same recall entirely on the phone. No clip, frame or still is shown here: the project's own policy blocks publishing footage until face and plate blurring and speaker consent are in place.
xPilot
Prototype
The core works end to end and can be demonstrated; it is not production-hardened or generally available.
A voice-first driving copilot for Indian roads that keeps working without a network connection.
Existing voice assistants for driving assume constant connectivity and don't handle India-specific driving needs well.
A tap-to-talk Android copilot that answers questions over a live map during a trip, using on-device rules and a local AI model first and falling back to a cloud AI model only when needed.
The local-first, cloud-fallback design is the point: it answers from on-device rules and a local model before ever calling out, so it degrades gracefully rather than failing outright when the network drops.
The cloud fallback carries a free-tier daily request limit, so the copilot is not positioned as unlimited cloud-dependent capability.
a Kotlin/Compose Android app running an on-device language model
