QazTender AI

Real AI + real web intelligence MVP for procurement/tender risk detection.

What this repo contains

• backend/ — FastAPI backend

  • Unified ingestion pipeline: text / file / URL → canonical text → stable chunks
  • Deterministic canonical_text_hash (debug mode)
  • Supervised ML baseline (TF-IDF + LogisticRegression) with explainable signals
  • Web intelligence: live market price discovery (no API keys required; best-effort) + caching
  • Consistency regression tests (same input ⇒ same hash & same score)

• frontend/ — Next.js UI (premium dark UI). Branding is QazTender AI.

• scripts/ — training utilities (prepare/train/evaluate)

Quickstart

Backend

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Endpoints

• POST /analyze/text
• POST /analyze/url
• POST /analyze/file
• POST /analyze (legacy structured payload)

Training (baseline)

From repo root:

PYTHONPATH=backend python scripts/train_model.py --data backend/app/ml/demo_data.jsonl
PYTHONPATH=backend python scripts/evaluate.py --data backend/app/ml/demo_data.jsonl

Tests

cd backend
pytest -q

Privacy

• No personal data collection is implemented.
• Minimal logs by default.
• Web market results are cached in SQLite for the demo.