🚀 Redrob AI

AI-Powered Candidate Ranking & Resume Intelligence Platform

Redrob AI helps recruiters and hiring managers instantly process massive datasets of resumes, ranking them with semantic understanding, skill matching, and explainable AI insights.

Instead of manually reading thousands of resumes, hiring teams can upload a candidate database and instantly explore the best fits, calculate custom weighted strategies, and view automated executive summaries through an interactive dashboard.

✨ Features

🔍 Semantic Vector Search

Rank candidates using contextual intent instead of keyword matching.

Example:

We need a Senior Python Developer with FastAPI and Docker experience who has built scalable APIs.

Redrob AI searches the candidate pool using vector similarity to find candidates with relevant work history and titles.

📊 Dynamic Ranking Playground

Adjust hiring weights in real-time. Move sliders to prioritize:

Qualification (Core Technical Skills + Experience + Education)
Compatibility (Semantic JD Overlap)
Behavior (Recruiter Engagement & Responsiveness)
Confidence (Data Completeness)

Leaderboards update instantly without invoking LLMs again.

🧠 AI Hiring Decision Copilot

Generate detailed executive summaries for candidates:

Match Score & Verdict: Star ratings and recommendation statuses.
Why Not Rank #1?: Automated competitive gap analysis detailing why a candidate didn't make the top spot.
Ramp-up Prediction: Immediate, 2-week, or 1-month onboarding estimations.
Strengths & Concerns: Bulleted business pros and cons.
Interview Focus Areas: Custom questions tailored to the candidate's exact resume gaps.

🏛 System Architecture

                    ┌───────────────┐
                    │ React Frontend│ (Vite + CSS Modules)
                    └───────┬───────┘
                            │
                            ▼
                    ┌───────────────┐
                    │ FastAPI Backend│ (Uvicorn)
                    └───────┬───────┘
                            │
        ┌───────────────────┼───────────────────┐
        ▼                   ▼                   ▼

 ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
 │ Resume      │    │ Vector      │    │ Intelligence│
 │ Ingestion   │    │ Database    │    │ Engine      │
 └──────┬──────┘    └──────┬──────┘    └──────┬──────┘
        │                  │                  │
        ▼                  ▼                  ▼

 ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
 │ Structured  │    │ FAISS       │    │ Gemini 2.5  │
 │ JSONL Parser│    │ Index       │    │ Flash API   │
 └─────────────┘    └─────────────┘    └─────────────┘

🛠 Tech Stack

Frontend

React 18
Vite
Recharts (Radar charts)
Lucide React (Icons)
CSS Modules (Scoped styling)

Backend

FastAPI (Python 3.11+)
Uvicorn
FAISS (Vector Database)
Sentence-Transformers (all-MiniLM-L6-v2)
Pydantic v2

AI Provider

Google GenAI SDK (Gemini 2.5 Flash)

📊 Platform Capabilities

Capability	Description
Resilient Extraction	Resilient dual-stage parsing (Regex Baseline -> Gemini Enhanced)
Instant Uploads	Skips neural network embeddings on restart if vectors already exist on disk
Vector Ranking	Evaluates 100,000 resumes in milliseconds using FAISS
Explainable AI	Detailed gap analysis ("Why Not Rank #1?") and ramp-up predictions
Recalculation	Vector-matrix dot products for real-time slider re-ranking
Export	CSV export of final ranked leaderboards

📁 Project Structure

INDIA-RUNS/
│
├── backend/
│   ├── app/
│   │   ├── api/          — Routes (upload, rank, recalculate, export)
│   │   ├── config/       — App settings
│   │   ├── models/       — Schemas and JSON formats
│   │   └── pipeline/     — Ingestion, embedding, FAISS vector store, ranking & explainability
│   │
│   └── requirements.txt
│
├── frontend/
│   ├── src/
│   │   ├── components/   — Results panel, setup panel, sliders, radar charts
│   │   ├── pages/        — Main dashboard layout
│   │   ├── lib/          — API caller and mock defaults
│   │   └── index.css     — Custom design tokens
│   │
│   └── package.json
│
└── README.md

🚀 Local Setup

Clone Repository

git clone https://github.com/WebDoveleprrr/Redrob-AI.git
cd Redrob-AI

Backend Setup

cd backend

python -m venv venv

# Windows
venv\Scripts\activate

# Linux / Mac
source venv/bin/activate

pip install -r requirements.txt

uvicorn app.main:app --reload

Backend URL:

http://localhost:8000

Frontend Setup

cd frontend

npm install

npm run dev

Frontend URL:

http://localhost:5173

🔧 Environment Variables

Backend (`backend/.env`)

GEMINI_API_KEY=your_google_ai_studio_api_key

🌐 API Endpoints

Candidate Ingestion & Database

Method	Endpoint	Description
POST	/api/v1/upload_db	Upload `candidates.jsonl` database (Cached vectors load instantly)

Ranking & Intelligence

Method	Endpoint	Description
POST	/api/v1/rank	Rank candidates against raw Job Description (Gemini Enhanced)
POST	/api/v1/recalculate	Fast vector re-ranking using customized slider weights
GET	/api/v1/export_csv	Export ranked candidates to CSV

📈 Future Improvements

Multi-recruiter team permissions
Custom weighted score presets saving
Dynamic JSONL resume schema support (auto-detecting resume fields)
Direct PDF/Word resume uploads with OCR parsing
Calendar integration for direct scheduling of top matches

👨‍💻 Author

Rohit Chowdary
Computer Science Engineering Student

Built to explore how hybrid search engines, vector indexing, and generative models can be combined to solve large-scale talent acquisition bottlenecks.

⭐ Why Redrob AI?

Redrob AI is not just another wrapper for an LLM.

It combines:

Fallback Resilience: Offline Regex parsing automatically boots up if Gemini API limits are hit.
Speed: Embeddings are computed only once, saved to a binary FAISS database, and reloaded instantly.
Mathematical Precision: Instead of asking an LLM to guess a ranking order, a vector dot product combines semantic match with behavioral scores to order candidates deterministically.
Explainability: Every ranking comes with concrete reasons, gap analysis, and tailored interview advice.

FilesExpand file tree

README.md

Latest commit

History