DocuChat — Document Q&A
Upload a PDF and chat with it using RAG-based retrieval
PDF chunking and embedding via Gemini Embedding 2 (2048-dim vectors)
RAG search over Firestore vector index for grounded answers
Streaming responses via SSE with source citations
Session history with resumable conversations
Mentor — Adaptive Tutor
Upload a textbook and get a structured course with curriculum, lessons, and quizzes
Ingestion pipeline: PDF parser → concept extraction → dependency graph → curriculum generation
Guided lessons that adapt to learner progress
Structured quizzes (MCQ + fill-in-the-blank) with weighted mastery (EMA)
Spaced repetition scheduling (SM-2 algorithm)
Knowledge graph visualization (force-directed 2D)
Full-course exam simulator
Illustra — Image Generation
Text-to-image with automatic prompt enhancement via Gemini
Multi-provider: Stability AI (default) and OpenAI GPT Image 2.0
Provider prefix routing ([openai], [openai][high]) for quick switching
Image upload to Google Cloud Storage with personal gallery
50 credits per image, prompt enhancement included
Platform & Architecture
Go backend with Google ADK v2 agent framework and Chi router
Next.js 16 frontend with React 19 and Tailwind CSS v4
Firebase Auth (Google + email/password) across both services
Cloud Firestore for data, Google Cloud Storage for files
Pay-per-token billing via Razorpay — 1 credit = ₹0.10, 100 free on signup
Deployed on Google Cloud Run (asia-south1) with Firebase Hosting CDN
CI/CD via GitHub Actions with staging and production branches
Two planned agents: Signal and Weaver
Tools & Technologies
Backend: Go, Chi v5, Google ADK v2
AI/LLM: Gemini 2.5 Flash, Gemini Embedding 2
Image Gen: Stability AI, OpenAI GPT Image 2.0
Frontend: Next.js 16, React 19, Tailwind CSS v4, Bun
Auth: Firebase Auth + Admin SDK
Database: Cloud Firestore
Storage: Google Cloud Storage
Payments: Razorpay
Infra: Google Cloud Run, Firebase Hosting, GitHub Actions
Linting: Biome, gofmt