Doc Doctor

• 

  home splash

• 

  workspace

• 

  Architecture Diagram

## Inspiration Doc Doctor grew out of late-night conversations with India-based counsel who spend hours flagging boilerplate risks that never reach partners’ desks. When Google announced the Agent Development Kit, I saw a chance to encode that institutional know-how into a transparent, explainable AI copilot that actually cites Indian statutes instead of vague “compliance concerns.” ## What it does The platform ingests an employment or vendor agreement, masks PII, orchestrates eight Gemini-powered specialists through Google ADK, applies ICA §27, Arbitration §20, and DPDP Act guardrails, and returns an executive summary plus clause-by-clause risks, citations, and rewrites inside a Vue workspace backed by Firestore and Cloud Storage.

How we built it

I containerized a FastAPI backend that mints signed upload URLs, triggers the ADK runner, and persists normalized outputs in Firestore. Vertex AI handles reasoning,Document AI extracts structure, and a Tailwind/Vite front end visualizes stage progress in real time. Docker images deploy to Cloud Run with environment-driven configuration.

Challenges we ran into

Getting ADK sessions to persist stage state reliably required custom tools, and aligning clause IDs across agents was brittle. Document AI rejects some DOCX mimetypes, so I had to fall back to raw decoding. Strict citation allowlists also meant curating an India-specific rulepack instead of relying on model guesses. ## Accomplishments that we're proud of The agent timeline mirrors backend orchestration perfectly, and every HIGH flag references an IndiaCode or MeitY source. I’m proud that the pipeline remains serverless-friendly—no long-lived workers, just stateless invocations with auditable Firestore trails.

What we learned

I learned how to tame ADK’s session primitives, how Gemini 2.5 behaves when forced into legal guardrails, and how much value hard-coded heuristics add when paired with LLM judgment. I also picked up deployment best practices for Cloud Run plus Document AI.

What's next for Doc Doctor

Next up: expand the rulepack to GDPR and Singapore PDPA, add Google Docs export hooks, strengthen citation validation with embeddings, and introduce a supervision console so lawyers can annotate AI findings and improve the agent prompts continuously.

Built With

• adk
• cloud
• css
• docker
• document
• fastapi
• firestore
• gemini
• pinia
• pub/sub
• pydantic
• python
• run
• storage
• tailwind
• typescript
• vertex
• vue