BidIQ

Responding to RFPs manually is time-consuming. You read through a long document, identify every requirement, figure out which ones your company meets, write a compliance response for each, and then hope your bid scores well enough to win. That whole process — for a single RFP — can easily take days.

BidIQ automates it. Upload the PDF, and it extracts every requirement, runs each one through a RAG pipeline against your historical bid library, generates draft proposal paragraphs for the ones you match, and scores your overall win probability using a trained ML model.

Three services: a Node.js/Express backend, a React frontend, and a Python FastAPI service for vector search and ML scoring.

What it does

Step	What happens
PDF Upload	Upload an RFP or tender PDF. Raw text is extracted and cleaned.
Requirement Extraction	One LLM call splits the doc into mandatory, technical, and financial requirements.
RAG Matching	Each technical requirement is embedded and queried against ChromaDB. Distance thresholds decide `matched` vs `gap`.
Classification + Fact-Check	Mandatory/financial requirements are classified as `fact` or `experience` in one batch call. Facts are checked against the company profile. Experience items go through RAG.
Draft Generation	Matched requirements get proposal paragraphs written by the LLM, using the matched capability as evidence.
Win Probability	Compliance %, gap count, sector, budget, and timing are fed into a scikit-learn logistic regression model.

Other things it has: JWT auth with httpOnly cookies, multi-workspace support, per-task LLM provider routing (OpenRouter / Hugging Face / Ollama), automatic provider failover.

System Architecture

%%{init: {"theme": "base", "themeVariables": {
  "background": "#0d1117",
  "primaryColor": "#161b22",
  "primaryTextColor": "#e6edf3",
  "primaryBorderColor": "#30363d",
  "lineColor": "#58a6ff",
  "secondaryColor": "#161b22",
  "tertiaryColor": "#21262d",
  "clusterBkg": "#161b22",
  "clusterBorder": "#388bfd",
  "edgeLabelBackground": "#0d1117",
  "fontFamily": "ui-monospace,SFMono-Regular,SF Mono,Menlo,monospace"
}}}%%
flowchart LR
    classDef feNode fill:#1d4ed8,stroke:#93c5fd,color:#fff,stroke-width:2px
    classDef beNode fill:#15803d,stroke:#86efac,color:#fff,stroke-width:2px
    classDef dbNode fill:#b91c1c,stroke:#fca5a5,color:#fff,stroke-width:2px
    classDef aiNode fill:#7c3aed,stroke:#c4b5fd,color:#fff,stroke-width:2px

    subgraph FE ["  Frontend  —  :5173  "]
        direction TB
        UI["WorkspacePage.jsx\nDashboardPage.jsx\nLoginPage.jsx"]
        Store["workspaceStore.js\ndocumentStore.js\nauthStore.js\n— Zustand v5"]
        UI --> Store
    end

    subgraph BE ["  Backend  —  Node.js / Express  —  :5000  "]
        direction TB
        Controllers["document.controller\nextraction.controller\nmatch.controller\ndraft.controller\nscore.controller\nauth.controller\ncompanyProfile.controller"]
        Utils["llm.js — multi-provider dispatcher\nrfpAnalysis.js — prompts + batch logic\npdfUtils.js — PDF text cleaning\nbidUtils.js — compliance + budget math\naiService.js — AI service HTTP client"]
        Controllers --> Utils
    end

    subgraph DB ["  MongoDB Atlas  "]
        Collections["Document · Extraction\nMatch · DraftSection\nUser · CompanyProfile · Capability"]
    end

    subgraph PY ["  Python AI Service  —  :8000  "]
        direction TB
        MatchEP["POST /match\nOllama nomic-embed-text\nChromaDB vector search"]
        PredictEP["POST /predict\nscikit-learn Logistic Regression\n+ StandardScaler"]
    end

    FE -- "HTTP REST / httpOnly cookies" --> BE
    BE -- "Mongoose / MongoDB Atlas" --> DB
    BE -- "HTTP" --> PY

    class UI,Store feNode
    class Controllers,Utils beNode
    class Collections dbNode
    class MatchEP,PredictEP aiNode

Tech Stack

Backend

Layer	Technology
Runtime	Node.js 22 (ESM modules)
Framework	Express.js v5
Database	MongoDB Atlas via Mongoose v9
Authentication	JWT (`jsonwebtoken`) + `bcryptjs` + `httpOnly` cookies
File Handling	Multer (PDF upload)
PDF Parsing	`pdf-parse`
Excel Reading	`xlsx` (capability library seeding)
Package Manager	pnpm

AI Service (Python)

Layer	Technology
Framework	FastAPI
Vector Store	ChromaDB (persistent local store)
Embeddings	Ollama `nomic-embed-text:latest`
Win Prediction	scikit-learn Logistic Regression + StandardScaler
Data Processing	pandas

Frontend

Layer	Technology
Framework	React 19 + Vite 8
Styling	TailwindCSS v4
State Management	Zustand v5
HTTP Client	Axios
Icons	Lucide React
Notifications	react-hot-toast
Routing	React Router DOM v7

LLM Providers

Task	Default Provider	Model
Extraction	OpenRouter	`inclusionai/ling-3.0-flash:free`
Matching	Hugging Face	`Qwen/Qwen2.5-7B-Instruct`
Draft	Hugging Face	`Qwen/Qwen2.5-7B-Instruct`
Scoring	OpenRouter	`inclusionai/ling-3.0-flash:free`

Project Structure

BidIQ/
├── backend/                        # Node.js / Express API
│   ├── src/
│   │   ├── app.js                  # Express app entry point
│   │   ├── config/
│   │   │   └── db.js               # MongoDB connection
│   │   ├── controllers/
│   │   │   ├── auth.controller.js
│   │   │   ├── companyProfile.controller.js
│   │   │   ├── document.controller.js
│   │   │   ├── draft.controller.js
│   │   │   ├── extraction.controller.js
│   │   │   ├── match.controller.js
│   │   │   └── score.controller.js
│   │   ├── middleware/
│   │   │   └── auth.middleware.js   # JWT protect middleware
│   │   ├── models/
│   │   │   ├── Capability.js
│   │   │   ├── CompanyProfile.js
│   │   │   ├── Document.js
│   │   │   ├── DraftSection.js
│   │   │   ├── Extraction.js
│   │   │   ├── Match.js
│   │   │   └── User.js
│   │   ├── routes/
│   │   │   ├── auth.routes.js
│   │   │   ├── companyProfile.routes.js
│   │   │   ├── document.routes.js
│   │   │   ├── draft.routes.js
│   │   │   ├── extraction.routes.js
│   │   │   ├── match.routes.js
│   │   │   └── score.routes.js
│   │   ├── scripts/
│   │   │   └── seedCapabilities.js  # Seeds MongoDB Capability collection from CSV
│   │   ├── data/
│   │   │   └── Capability_Library.xlsx
│   │   └── utils/
│   │       ├── aiService.js         # HTTP client to Python AI service
│   │       ├── bidUtils.js          # Compliance % calculation + budget parser
│   │       ├── llm.js               # LLM provider dispatcher with failover
│   │       ├── pdfUtils.js          # PDF text cleaning and normalization
│   │       └── rfpAnalysis.js       # All prompts, batch classification, fact-check, draft logic
│   ├── uploads/                     # Multer uploaded PDFs (auto-created)
│   ├── .env                         # Local secrets (gitignored)
│   ├── .env.example                 # Environment variable template
│   └── package.json
│
├── frontend/                        # React + Vite SPA
│   ├── public/
│   │   └── favicon.png
│   ├── src/
│   │   ├── components/
│   │   │   ├── ComplianceMatrixTable.jsx
│   │   │   ├── DraftSectionCard.jsx
│   │   │   ├── ExtractedDetailsView.jsx
│   │   │   ├── Navbar.jsx
│   │   │   ├── ProtectedRoute.jsx
│   │   │   ├── RequirementCard.jsx
│   │   │   ├── SkeletonLoader.jsx
│   │   │   ├── UploadModal.jsx
│   │   │   └── WinProbabilityCard.jsx
│   │   ├── lib/
│   │   │   └── api.js               # Axios instance with base URL + credentials
│   │   ├── pages/
│   │   │   ├── DashboardPage.jsx    # RFP document repository
│   │   │   ├── LoginPage.jsx
│   │   │   ├── SignupPage.jsx
│   │   │   └── WorkspacePage.jsx    # Full RFP workspace
│   │   ├── stores/
│   │   │   ├── authStore.js         # Auth state (Zustand)
│   │   │   ├── documentStore.js     # Document list state
│   │   │   └── workspaceStore.js    # Workspace pipeline state
│   │   ├── App.jsx
│   │   ├── index.css
│   │   └── main.jsx
│   ├── index.html
│   └── package.json
│
├── ai-service/                      # Python FastAPI AI service
│   ├── main.py                      # /match and /predict endpoints
│   ├── notebooks/
│   │   └── BidIQ.ipynb              # Model training and exploration notebook
│   ├── models/
│   │   ├── logistic_model.pkl       # Trained win probability model
│   │   └── scaler.pkl               # StandardScaler for model input
│   ├── chroma_store/                # Persistent ChromaDB vector store
│   ├── scripts/                     # Training/seeding scripts
│   ├── requirements.txt
│   └── venv/
│
└── README.md

Getting Started

Prerequisites

Requirement	Details
Node.js	v22+ (ESM support required)
pnpm	`npm install -g pnpm`
Python	3.10+
Ollama	Installed and running locally (ollama.com)
MongoDB Atlas	Free cluster at cloud.mongodb.com
OpenRouter Account	Free API key at openrouter.ai
Hugging Face Account	Free API key at huggingface.co

Required Ollama models (for vector embeddings, used by the AI service):

ollama pull nomic-embed-text:latest

Optionally, for local LLM inference:

ollama pull phi4-mini:latest

1. Backend Setup

cd backend
pnpm install

# Copy and populate environment variables
cp .env.example .env
# Edit .env with your credentials

# Seed metadata to MongoDB (requires MongoDB running/URI configured)
node src/scripts/seedCapabilities.js

# Start development server
pnpm dev
# Server runs on http://localhost:5000

2. AI Service Setup

Ensure Ollama is running and has the embedding model downloaded:

ollama serve
ollama pull nomic-embed-text:latest

Then configure and start the Python AI Service:

cd ai-service

# Create virtual environment
python -m venv venv

# Activate (Windows)
venv\Scripts\activate

# Activate (macOS/Linux)
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Seed the ChromaDB local vector store (Ollama must be running)
python scripts/seed_chroma.py

# Start the AI service
uvicorn main:app --host 0.0.0.0 --port 8000 --reload
# AI service runs on http://localhost:8000

3. Frontend Setup

cd frontend
pnpm install

# Start development server
pnpm dev
# App runs on http://localhost:5173

Environment Variables

`backend/.env.example`

# ── Server ─────────────────────────────────────────────────
PORT=5000

# ── Database ───────────────────────────────────────────────
MONGO_URI=mongodb+srv://<user>:<password>@<cluster>.mongodb.net/bidiq?appName=Cluster0

# ── Authentication ─────────────────────────────────────────
JWT_SECRET=your_super_secret_jwt_key_here
CLIENT_URL=http://localhost:5173

# ── LLM Provider (default fallback if task-specific not set) ─
# Options: openrouter | huggingface | ollama
LLM_PROVIDER=openrouter

# ── Per-Task LLM Provider Routing ──────────────────────────
# Each task can independently use a different LLM provider.
# Options per task: openrouter | huggingface | ollama
EXTRACTION_LLM_PROVIDER=openrouter
MATCH_LLM_PROVIDER=huggingface
DRAFT_LLM_PROVIDER=huggingface
SCORE_LLM_PROVIDER=openrouter

# ── OpenRouter ──────────────────────────────────────────────
OPENROUTER_API_KEY=sk-or-v1-your-key-here
OPENROUTER_MODEL=inclusionai/ling-3.0-flash:free

# ── Hugging Face ────────────────────────────────────────────
HF_API_KEY=hf_your_key_here
HF_ROUTER_URL=https://router.huggingface.co/v1/chat/completions
HF_INFERENCE_URL=https://api-inference.huggingface.co/models
HF_MODEL_DEFAULT=Qwen/Qwen2.5-7B-Instruct

# ── Ollama (Local) ──────────────────────────────────────────
OLLAMA_MODEL=phi4-mini:latest
OLLAMA_HOST=http://127.0.0.1:11434

API Reference

Authentication

Method	Endpoint	Description
`POST`	`/api/auth/signup`	Register a new user
`POST`	`/api/auth/login`	Login and receive session cookie
`POST`	`/api/auth/logout`	Clear session cookie
`GET`	`/api/auth/me`	Get current authenticated user

Documents (protected)

Method	Endpoint	Description
`GET`	`/api/documents`	List all uploaded RFP documents
`POST`	`/api/documents/upload`	Upload a new RFP PDF (`multipart/form-data`, field: `rfpFile`)
`GET`	`/api/documents/:id/workspace`	Fetch full workspace state (document, extraction, matches, drafts)
`POST`	`/api/documents/:id/extract`	Run LLM extraction on a document

Extractions (protected)

Method	Endpoint	Description
`POST`	`/api/extractions/:id/match`	Run RAG + fact-check matching on an extraction
`POST`	`/api/extractions/:id/draft`	Generate proposal draft sections
`GET`	`/api/extractions/:id/score`	Calculate compliance score + win probability

Company Profile (protected)

Method	Endpoint	Description
`POST`	`/api/company-profile`	Create or update the company profile

AI Service (port 8000)

Method	Endpoint	Description
`POST`	`/match`	Vector similarity match for a requirement text
`POST`	`/predict`	Win probability prediction using logistic regression
`GET`	`/docs`	FastAPI interactive Swagger docs

LLM Provider Routing

Each pipeline task (extraction, matching, drafting, scoring) independently routes to a configured LLM provider via env vars. If the primary provider fails, it automatically falls back through the chain.

Failover Chain

Primary Provider	Failover Order
`openrouter`	OpenRouter → Hugging Face
`huggingface`	Hugging Face → OpenRouter
`ollama`	Ollama → Hugging Face → OpenRouter

If a provider returns an error or times out, the next one in the chain is tried silently.

Model Recommendations

Task	Recommended Free Model
Extraction (long-context)	`inclusionai/ling-3.0-flash:free` (OpenRouter)
Batch Matching	`Qwen/Qwen2.5-7B-Instruct` (Hugging Face)
Draft Generation	`Qwen/Qwen2.5-7B-Instruct` (Hugging Face)
Sector Scoring	`inclusionai/ling-3.0-flash:free` (OpenRouter)

AI Pipeline Overview

%%{init: {"theme": "base", "themeVariables": {
  "background": "#0d1117",
  "primaryColor": "#161b22",
  "primaryTextColor": "#e6edf3",
  "primaryBorderColor": "#30363d",
  "lineColor": "#58a6ff",
  "secondaryColor": "#161b22",
  "tertiaryColor": "#21262d",
  "clusterBkg": "#161b22",
  "clusterBorder": "#388bfd",
  "edgeLabelBackground": "#0d1117",
  "fontFamily": "ui-monospace,SFMono-Regular,SF Mono,Menlo,monospace"
}}}%%
flowchart TD
    classDef ingest   fill:#0369a1,stroke:#7dd3fc,color:#fff,stroke-width:2px
    classDef extract  fill:#b45309,stroke:#fcd34d,color:#fff,stroke-width:2px
    classDef rag      fill:#15803d,stroke:#86efac,color:#fff,stroke-width:2px
    classDef llmcall  fill:#7c3aed,stroke:#c4b5fd,color:#fff,stroke-width:2px
    classDef factchk  fill:#be185d,stroke:#f9a8d4,color:#fff,stroke-width:2px
    classDef draft    fill:#0f766e,stroke:#5eead4,color:#fff,stroke-width:2px
    classDef scoring  fill:#b91c1c,stroke:#fca5a5,color:#fff,stroke-width:2px
    classDef split    fill:#334155,stroke:#94a3b8,color:#e2e8f0,stroke-width:1px

    PDF(["PDF Upload"])
    Clean["pdf-parse + pdfUtils.js\nClean and normalize raw text"]
    PDF --> Clean
    Clean --> Extract

    subgraph Extract [" Step 1 — Requirement Extraction  ·  1 LLM call — OpenRouter: ling-3.0-flash "]
        direction LR
        ExtractNode["extraction.controller.js  ·  buildExtractionPrompt\n\nSaved to MongoDB Extraction:\n  mandatoryRequirements[] · technicalRequirements[]\n  financialRequirements[] · deliverables[]\n  title · org · deadline · budget · rfpNumber"]
    end

    Extract --> TechReqs["technicalRequirements[]"]
    Extract --> MandFin["mandatoryRequirements[]\nfinancialRequirements[]"]

    subgraph TechMatch [" Step 2a — RAG Matching  per technical requirement "]
        TechRAG["aiService.js  POST :8000/match\nOllama nomic-embed-text embedding\nChromaDB cosine similarity search\n\ndistance less than 350  matched\ndistance greater than 400  gap"]
    end

    TechReqs --> TechMatch

    subgraph ClassifyStep [" Step 2b — Batch Classification  ·  1 LLM call — HuggingFace: Qwen2.5-7B "]
        ClassifyNode["classifyRequirementsBatch\n\nfact    verifiable company attribute\nexperience    past project evidence required"]
    end

    MandFin --> ClassifyStep
    ClassifyStep --> ExpItems["Experience items"]
    ClassifyStep --> FactItems["Fact items"]

    subgraph ExpMatch [" Step 2c — RAG Matching  experience items "]
        ExpRAG["POST :8000/match\nSame vector search pipeline\nmatched or gap"]
    end

    subgraph FactCheck [" Step 2d — Batch Fact-Check  ·  1 LLM call — HuggingFace: Qwen2.5-7B "]
        FactNode["factCheckRequirementsBatch\nvs CompanyProfile document\n\nPASS · FAIL · INSUFFICIENT_DATA"]
    end

    ExpItems --> ExpMatch
    FactItems --> FactCheck

    TechMatch --> Drafts
    ExpMatch --> Drafts
    FactCheck --> Drafts

    subgraph Drafts [" Step 3 — Batch Draft Generation  ·  1 LLM call — HuggingFace: Qwen2.5-7B "]
        DraftNode["generateDraftsBatch\nMatched  evidence-based proposal paragraphs\nPassed facts  compliance confirmation sentences\nSaved to MongoDB DraftSection collection"]
    end

    Drafts --> Scoring

    subgraph Scoring [" Step 4 — Bid Scoring "]
        direction LR
        Compliance["bidUtils.js  computeBidStats\ncompliance_percent · gaps_found · doc_pages"]
        Sector["classifySector  1 LLM call — OpenRouter\nEducation · IT Services · Healthcare · Finance..."]
        Win["POST :8000/predict\nscikit-learn LogisticRegression\nfeatures: budget · response_time · compliance\ngaps · sector · submission month\n\nOutput: Win probability 0 to 1"]
        Compliance --> Win
        Sector --> Win
    end

    class PDF,Clean ingest
    class ExtractNode extract
    class TechRAG,ExpRAG rag
    class ClassifyNode llmcall
    class FactNode factchk
    class DraftNode draft
    class Compliance,Sector,Win scoring
    class TechReqs,MandFin,ExpItems,FactItems split

Total LLM calls per full RFP analysis: ~4–6 calls (down from 30–40 sequential calls)

Data Models

Document

{ originalName, filePath, extractedText, pageCount, uploadedAt }

Extraction

{
  document,                    // ref: Document
  title, organization, rfpNumber, country,
  submissionDeadline, projectDuration, contractType, estimatedBudget,
  mandatoryRequirements[],     // array of strings
  technicalRequirements[],
  financialRequirements[],
  deliverables[],
  requiredDocuments[],
  evaluationCriteria[],
  contact: { email, address },
  rawLLMResponse
}

Match

{
  extraction,                  // ref: Extraction
  requirementText,
  requirementType,             // technical | mandatory | financial
  method,                      // rag | fact_check
  status,                      // matched | gap | pass | fail | insufficient_data
  matchedCapabilities[],       // [{ capId, distance, documentText }]
  factCheckResult: { verdict, reason }
}

DraftSection

{
  extraction,                  // ref: Extraction
  requirementText,
  draftText,
  basedOnCapability,           // ref: Capability
  source                       // rag | fact_check
}

CompanyProfile

{
  name, registrationYear, country, certifications[],
  annualTurnover, sectors[], pastProjects[],
  blacklisted, officeLocations[]
}

Data Seeding

Both databases need to be seeded before the app is functional. The source data is Capability_Library.csv — a set of historical bid records used for RAG matching and win probability training.

1. MongoDB — Bid History Metadata

Script: backend/src/scripts/seedCapabilities.js
Run from: backend/ directory
```
node src/scripts/seedCapabilities.js
```
Clears the existing Capability collection and re-inserts all records from Capability_Library.csv (budget, sector, outcome, compliance %, etc.).

2. ChromaDB — RAG Vector Store

Script: ai-service/scripts/seed_chroma.py
Run from: ai-service/ with virtualenv active
```
python scripts/seed_chroma.py
```
Drops the existing ChromaDB collection, re-embeds all bid records using nomic-embed-text via Ollama, and stores them locally under chroma_store/.
Ollama must be running before this script is executed.

License

MIT

Name		Name	Last commit message	Last commit date
ai-service		ai-service
backend		backend
docs		docs
frontend		frontend
.gitignore		.gitignore
README.md		README.md

Latest commit

History

Folders and files

Repository files navigation