TrackIt‑AI

This project turns receipt images into structured expense data, stores it in Supabase Postgres, and lets users ask natural‑language questions about their spending — now also accessible via Telegram.

---

Stack / Technologies

Layer	Tech
LLM (receipt parsing & categorization)	meta‑llama/llama‑4 models API
Text‑to‑SQL	sqlcoder-7b-2 via Cloudflare AI (HuggingFace: defog/sqlcoder-7b-2)
OCR	Mistral OCR (fallback: tesseract)
Backend API	Python + FastAPI + asyncio
Database	Supabase Postgres
Frontend	Next.js + Tailwind / shadcn
Messaging	Telegram Bot integration via python‑telegram‑bot

---

Demo Images

---

API Routes

Method	Path	Purpose
GET	/receipts/user/{user_id}	list receipts by user
POST	/receipts/extract	extract data only (preview)
POST	/receipts/save	save structured receipt JSON
POST	/receipts/telegram_upload	OCR + LLM extract → save via Telegram bot
POST	/query/ask	NL query → SQLCoder → execute → summary

---

Local Setup (dev)

git clone https://github.com/your-handle/trackit-ai
d
# ---------- backend ----------
cd trackit-ai/backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload   # http://localhost:8000

# ---------- frontend ----------
cd ../frontend
npm install
npm run dev               # http://localhost:3000

# ---------- telegram‑bot ----------
cd ../telegram-bot
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python main.py            # runs the Telegram bot

Env Variables

backend/.env

GROQ_API_KEY=your_groq_key
GEMINI_API_KEY=your_gemini_key
SUPABASE_URL=https://xyz.supabase.co
SUPABASE_KEY=service_role_or_anon_key
PW_SALT=random_string

frontend/.env.local

NEXT_PUBLIC_API_URL=http://localhost:8000

telegram-bot/.env

TELEGRAM_BOT_TOKEN=your_bot_token