Skip to content

VishakBaddur/procure-ai

BranchesTags

Repository files navigation

Procure AI

An AI-powered procurement intelligence platform for procurement officers to manage vendors, compare quotes, analyze legal agreements, research vendor reputation, and calculate Total Cost of Ownership — organized around Projects for multi-vendor, multi-document workflows.

Live demo: https://procure-ai-byk5.onrender.com

Features

AI Agents

  • Price Comparison Agent — Extracts and compares pricing from vendor quotes (PDF, image, text)
  • Legal Analysis Agent — Analyzes vendor agreements for risks, terms, and recommendations
  • Vendor Research Agent — Reputation scoring, red flag detection, sourced recommendations via Groq AI
  • TCO Agent — 5-year Total Cost of Ownership projections including hidden costs
  • Decision Agent — Final vendor recommendation with reasoning across all collected signals
  • Email Agent — Fetches and processes vendor quotes directly from email (IMAP/POP)
  • Embedding Agent — Chunks and embeds all vendor documents using Groq nomic-embed-text-v1.5 (768d)

Core Features

  • Project-based workflow — All procurement activity organized under Projects; each project tracks multiple vendors
  • JWT authentication — Secure register/login with bcrypt password hashing and 7-day tokens
  • Semantic search — Natural language search across all vendor documents using pgvector cosine similarity with HNSW index
  • Auto-embedding — Every uploaded document is automatically chunked and embedded in the background
  • Quote uploads — PDF, image, text per vendor; AI extracts and structures pricing
  • Agreement uploads — Legal agreements per vendor; AI scores risk and surfaces key terms
  • TCO analysis — 5-year cost projections
  • What-if analysis — Model cost scenarios with adjustable assumptions
  • Decision assistance — AI-generated final vendor recommendation
  • Analytics dashboard — Side-by-side comparisons across quotes, agreements, reviews, and TCO
  • Email integration (optional) — Connect IMAP/POP mailbox to auto-fetch vendor quotes

Tech Stack

Backend

  • FastAPI — Python web framework
  • PostgreSQL — Production database via SQLAlchemy ORM
  • pgvector — Vector similarity search extension (HNSW index, cosine distance)
  • Groq AI — Primary LLM for all agents + nomic-embed-text-v1.5 embeddings
  • pdfplumber — PDF text extraction
  • pytesseract + Pillow — OCR for image-based quotes
  • python-docx — Word document support
  • JWT (python-jose + bcrypt) — Authentication
  • python-dotenv — Environment variable management

Frontend

  • React 18 — UI framework
  • Vite — Build tool
  • React Router v6 — Navigation with protected routes
  • Recharts — Data visualization
  • Tailwind CSS — Styling
  • Radix UI — Accessible component primitives
  • Lucide React — Icons

Infrastructure

  • Render — Backend + frontend deployment (single Docker container)
  • Render PostgreSQL — Managed PostgreSQL with pgvector

Architecture

User → React Frontend (Vite/Tailwind)
          ↓ JWT-authenticated requests
FastAPI Backend
          ↓
┌─────────────────────────────────────┐
│  Agents (Groq AI)                   │
│  - Price Comparison                 │
│  - Legal Analysis                   │
│  - Vendor Research                  │
│  - TCO                              │
│  - Decision                         │
│  - Embedding (nomic-embed-text)     │
└─────────────────────────────────────┘
          ↓
PostgreSQL + pgvector
- Projects / Vendors / Documents
- Parsed data (JSON)
- Document embeddings (768d, HNSW index)
- Users (bcrypt hashed)

Project Structure

procure ai cursor/
├── backend/
│   ├── agents/
│   │   ├── price_comparison_agent.py
│   │   ├── legal_analysis_agent.py
│   │   ├── vendor_research_agent.py
│   │   ├── tco_agent.py
│   │   ├── decision_agent.py
│   │   ├── email_agent.py
│   │   └── embedding_agent.py
│   ├── database.py          # SQLAlchemy ORM + pgvector + auth helpers
│   ├── main.py              # FastAPI app, all endpoints
│   ├── models.py
│   ├── requirements.txt
│   └── uploads/
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   │   ├── AuthPage.jsx
│   │   │   ├── SemanticSearch.jsx
│   │   │   ├── ProjectList.jsx
│   │   │   ├── CreateProject.jsx
│   │   │   ├── ProjectDashboard.jsx
│   │   │   ├── ProjectLayout.jsx
│   │   │   ├── Dashboard.jsx
│   │   │   ├── Questionnaire.jsx
│   │   │   ├── UploadQuotes.jsx
│   │   │   ├── UploadAgreements.jsx
│   │   │   ├── QuotationComparison.jsx
│   │   │   ├── AgreementsComparison.jsx
│   │   │   ├── ReviewsComparison.jsx
│   │   │   ├── TCOComparison.jsx
│   │   │   ├── DecisionAssistance.jsx
│   │   │   └── WhatIfAnalysis.jsx
│   │   ├── AuthContext.jsx
│   │   ├── config.js
│   │   ├── App.jsx
│   │   └── main.jsx
│   ├── package.json
│   └── vite.config.js
├── docker-compose.yml
├── Dockerfile
├── start_backend.sh
├── start_frontend.sh
└── README.md

Setup Instructions

Prerequisites

Install pgvector (macOS)

git clone --branch v0.8.2 https://github.com/pgvector/pgvector.git /tmp/pgvector
cd /tmp/pgvector
make PG_CONFIG=/opt/homebrew/opt/postgresql@14/bin/pg_config
make install PG_CONFIG=/opt/homebrew/opt/postgresql@14/bin/pg_config

Backend Setup

cd backend
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Create a .env file in backend/:

GROQ_API_KEY=your_groq_api_key_here
DATABASE_URL=postgresql://localhost/procureai
SECRET_KEY=your_random_secret_key

# Optional
GEMINI_API_KEY=your_gemini_api_key
SERPAPI_KEY=your_serpapi_key
EMAIL_ADDRESS=your_email@example.com
EMAIL_PASSWORD=your_password
EMAIL_IMAP_SERVER=imap.gmail.com

Start the backend:

python3 main.py

API available at http://localhost:8000

Frontend Setup

cd frontend
npm install
npm run dev

Frontend available at http://localhost:3000

API Endpoints

Auth

Method Endpoint Description
POST /api/auth/register Register a new user
POST /api/auth/login Login and get JWT token
GET /api/auth/me Get current user

Search

Method Endpoint Description
POST /api/search Semantic search across vendor documents

Projects

Method Endpoint Description
POST /api/projects Create a new project
GET /api/projects List all projects
GET /api/projects/{project_id} Get project details
DELETE /api/projects/{project_id} Delete a project
GET /api/projects/{project_id}/dashboard Get project dashboard

Vendors

Method Endpoint Description
GET /api/projects/{project_id}/vendors List vendors
POST /api/projects/{project_id}/vendors Add a vendor
DELETE /api/projects/{project_id}/vendors/{vendor_id} Remove a vendor

Quotes and Agreements

Method Endpoint Description
POST /api/projects/{project_id}/vendors/{vendor_id}/quotations Upload a quote
GET /api/projects/{project_id}/quotations/comparison Compare all quotes
POST /api/projects/{project_id}/vendors/{vendor_id}/agreements Upload an agreement
GET /api/projects/{project_id}/agreements/comparison Compare all agreements

Research and Analysis

Method Endpoint Description
POST /api/projects/{project_id}/vendors/{vendor_id}/research Research a vendor
GET /api/projects/{project_id}/reviews/comparison Compare vendor reviews
GET /api/projects/{project_id}/tco/comparison TCO comparison
GET /api/projects/{project_id}/recommendation Get AI recommendation
POST /api/projects/{project_id}/recommendation/export Export recommendation
POST /api/projects/{project_id}/what-if Run what-if analysis

Email

Method Endpoint Description
POST /api/projects/{project_id}/email/fetch Fetch quotes from email
POST /api/projects/{project_id}/email/process Process fetched emails

Semantic Search

Every uploaded vendor document is automatically chunked (512-word chunks, 64-word overlap) and embedded using Groq's nomic-embed-text-v1.5 model (768 dimensions). Embeddings are stored in PostgreSQL via pgvector with an HNSW index (m=16, ef_construction=64) for sub-millisecond cosine similarity search at scale.

Query examples:

  • "vendors with overage fees"
  • "warranty terms longer than 12 months"
  • "SLA penalties for downtime"
  • "net payment terms"
  • "auto-renewal clauses"

Deployment

Deployed on Render as a single Docker container (frontend + backend).

Environment variables required on Render:

DATABASE_URL=<render-internal-postgres-url>
SECRET_KEY=<random-hex-string>
GROQ_API_KEY=<your-groq-key>
SERPAPI_KEY=<your-serpapi-key>  # optional

Future Enhancements

  • Multi-organization support with role-based access control
  • Expanded export formats (PDF reports, Excel)
  • Email auto-polling with threading
  • Mobile-responsive UI improvements
  • Confidence scoring on extracted fields with human review queue

License

MIT

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages