TenderSort AI-Powered Government Tender Evaluation System by Byte Verdict
Live Links ServiceURLFrontend (Netlify) https://tendersorttv2.netlify.app/ Backend API (Render) https://tendersortv2.onrender.com API Health Check https://tendersortv2.onrender.com/health Note: The backend is hosted on Render's free tier. If the health check takes 30-60 seconds to respond on first visit, that is normal — it is waking up from inactivity. Once awake, it runs at full speed.
What is TenderSort? TenderSort is an AI-powered web application that automates the evaluation of government tender bid submissions. In the traditional process, government officials manually read through hundreds of pages of bidder documents to check eligibility criteria — a process that takes days and is prone to human error and bias. TenderSort solves this by:
Automatically extracting eligibility criteria from tender documents using AI Evaluating each bidder's submission against those criteria Giving instant verdicts: Eligible, Not Eligible, or Manual Review Providing confidence scores and evidence for every decision Generating downloadable evaluation reports
Features
PDF Upload — Upload tender document and multiple bidder submissions at once AI Criteria Extraction — Automatically identifies eligibility criteria from tender PDF Automated Evaluation — Each bidder checked against all criteria with evidence Live Dashboard — Visual breakdown with donut chart and detailed table Manual Review Panel — Human review interface for low-confidence cases PDF Report Export — Download complete evaluation report as PDF Confidence Scoring — AI confidence percentage for every decision Filter and Search — Filter bidders by Eligible / Not Eligible / Needs Review
Tech Stack Frontend TechnologyPurposeReact.jsUI frameworkViteBuild toolTailwind CSSStylingReact RouterPage navigationRechartsDonut chart visualizationlocalStorageTemporary result storageNetlifyDeployment Backend TechnologyPurposePythonProgramming languageFastAPIWeb framework / API endpointsUvicornASGI serverPyMuPDF (fitz)PDF text extractionpython-multipartFile upload handlingpython-dotenvEnvironment variable managementrequestsHTTP calls to AI APIProcfileRender deployment configRenderCloud deployment AI / API TechnologyPurposeGroq APIAI inference platformLLaMA 3.3 70BLanguage model for evaluationMeta LLaMAOpen source AI model by Meta
System Architecture User uploads PDFs (Frontend - Netlify) | FastAPI receives files (Backend - Render) | PyMuPDF extracts text from PDFs | Groq API (LLaMA 3.3 70B) analyzes text Step 1: Extract eligibility criteria from tender Step 2: Evaluate each bidder against criteria | Returns verdict for each bidder: Eligible | Not Eligible | Manual Review | Frontend displays results on Dashboard
Getting Started Locally Prerequisites
Node.js v18+ Python 3.10+ Groq API key (free at console.groq.com)
- Clone the repository git clone https://github.com/zaidkamar08/tenderSortV2.git cd tenderSortV2
- Setup Backend cd backend pip install -r requirements.txt Create a .env file inside backend/: GROQ_API_KEY=your_groq_api_key_here Start the backend: uvicorn main:app --reload Backend runs at: http://localhost:8000
- Setup Frontend cd .. npm install Update the API URL in src/pages/Upload.jsx: fetch('http://localhost:8000/full-evaluation' Start the frontend: npm run dev Frontend runs at: http://localhost:5173
Project Structure
tenderSortV2/
├── backend/
│ ├── main.py
│ ├── requirements.txt
│ ├── Procfile
│ └── .env
├── src/
│ ├── pages/
│ │ ├── Home.jsx
│ │ ├── Upload.jsx
│ │ ├── Dashboard.jsx
│ │ ├── Review.jsx
│ │ └── Report.jsx
│ ├── components/
│ │ └── Navbar.jsx
│ └── data/
│ └── mockData.js
├── public/
├── package.json
├── index.html
└── README.md
How Evaluation Works
Upload Phase — User uploads 1 tender PDF and 1 or more bidder PDFs Criteria Extraction — AI reads the tender and identifies all eligibility criteria Bidder Evaluation — For each bidder, AI checks every criterion and returns pass/fail/review status, value found, confidence score, and reason Overall Verdict — If any mandatory criterion fails the bidder is Rejected. If all pass the bidder is Eligible. If confidence is low the case goes to Manual Review
Demo Test Files FileExpected ResultCRPF_Tender_Construction.pdfTender documentBharatInfrastructure_BidSubmission.pdfEligibleGreenfield_Builders_BidSubmission.pdfEligibleMetroWorks_BidSubmission.pdfNot EligibleHorizon_Contractors_BidSubmission.pdfNot EligibleAllied_Construction_BidSubmission.pdfManual Review For best demo: Upload CRPF tender + BharatInfrastructure + MetroWorks + Allied to get 1 Eligible, 1 Not Eligible, 1 Manual Review result.
Known Limitations
Free tier Groq API has rate limits — upload max 3 bidders at a time Render free tier backend sleeps after 15 minutes of inactivity with 30-60 second cold start Results stored in browser localStorage — cleared on browser data wipe No persistent database in current version (Supabase integration planned)
Future Improvements
Supabase database for persistent storage User authentication and multi-user support Support for scanned PDFs via OCR Email notifications for evaluation completion Audit trail and decision history
Team Byte Verdict Built for hackathon submission Full stack: React + FastAPI + LLaMA AI
License This project is built for educational and hackathon purposes. Built by Byte Verdict | Powered by LLaMA 3.3 70B via GroqSonnet 4.6Adaptive