cogniCV — AI-Powered Talent Screening Platform

cogniCV is an intelligent recruitment tool designed to automate and enhance the candidate screening process. Built for high-volume recruitment, it leverages Google's Gemini AI to parse resumes, analyze candidate fit against job descriptions, and provide explainable ranking shortlists.

Project Overview

cogniCV solves the challenge of manual resume screening by providing:

• AI-Powered Resume Parsing: Automatically extracts structured data (skills, experience, education) from PDF resumes.
• Intelligent Job Creation: An AI chat assistant that helps recruiters generate detailed, markdown-formatted job descriptions.
• Weighted Scoring Engine: Ranks candidates based on a multi-dimensional scoring model (Skills, Experience, Education, Relevance).
• Explainable AI: Provides specific "Strengths", "Gaps", and "Recommendations" for every candidate.
• Real-time Progress Tracking: Visual feedback during the AI screening process.
• Secure Sharing: Generate password-protected links to share candidate analysis with stakeholders.

Tech Stack

• Frontend: Next.js 15+, Tailwind CSS, Shadcn/UI, TanStack Query.
• Backend: Node.js, Express, TypeScript.
• Database: MongoDB (Mongoose).
• Messaging/Queue: RabbitMQ.
• Storage: Cloudflare R2 / MinIO (S3 Compatible).
• AI Engine: Google Gemini 2.5 Flash & Google gemini-embedding-001.

Links

• Live link: cogniCV
• Docs link: docs - Username : admin , Pass : admin

System Architecture

graph TD
    subgraph Client_Layer [Client Layer]
        A[Next.js Dashboard] -->|API Requests| B[Express API]
        A -->|Real-time Polling| B
    end

    subgraph Server_Layer [Server Layer]
        B -->|Auth & CRUD| C[(MongoDB)]
        B -->|Trigger Screening| D{Gemini AI}
        B -->|Queue Parsing| E[RabbitMQ]
    end

    subgraph Background_Layer [Background Processing]
        E -->|Ingestion Jobs| F[Worker Process]
        F -->|Fetch Files| G[Object Storage R2/MinIO]
        F -->|Extract Text| D
        F -->|Update Profiles| C
    end

    subgraph AI_Engine [AI Engine]
        D -->|JSON Mapping| B
        D -->|Batch Ranking| B
    end

Loading

1. Client (Next.js): Handles the dashboard, job management, and real-time visualization of screening results.
2. Server (Express): Orchestrates API requests, manages the database, and interacts with Gemini.
3. RabbitMQ: Manages asynchronous tasks like heavy resume parsing and ingestion.
4. Worker: Consumes messages from RabbitMQ to process resumes in the background.
5. Gemini API: Acts as the "Brain" for both resume parsing and candidate evaluation.
6. Object Storage: Stores uploaded resume files securely.

Setup Instructions

Prerequisites

• Node.js (v18+)
• pnpm (recommended) or npm/yarn
• Docker and Docker Compose
• Google Gemini API Key

1. Clone and Install

# Clone the repository
git clone <repo-url>
cd cognicv

# Install dependencies for both client and server
cd client && pnpm install
cd ../server && pnpm install

2. Infrastructure (Docker)

Run the required infrastructure services (RabbitMQ and MinIO):

docker-compose up -d

This will start RabbitMQ on port 5672/15672 and MinIO on port 9000/9001.

3. Environment Variables

You need to set up .env files in both the client and server directories.

Server (server/.env)

Copy server/.env.example to server/.env and fill in the values:

• MONGO_URI: Your MongoDB connection string.
• GEMINI_API_KEY: Your Google AI Studio API Key.
• RABBITMQ_URL: amqp://localhost (if using Docker).
• R2_ACCESS_KEY_ID, R2_SECRET_ACCESS_KEY: For storage (or local MinIO credentials).

Client (client/.env)

Copy client/.env.example to client/.env:

• NEXT_PUBLIC_SERVER_URL: http://localhost:8001 (default backend port).

4. Running the Application

Start the Backend:

cd server
pnpm dev

Start the Frontend:

cd client
pnpm dev

The application will be available at http://localhost:3000.

Environment Variables Detail

Backend (Server)

Variable	Description	Default
MONGO_URI	MongoDB Connection String	mongodb://localhost:27017/cognicv
SERVER_PORT	Port for the backend server	8001
GEMINI_API_KEY	Google AI API Key	(Required)
RABBITMQ_URL	RabbitMQ Connection String	amqp://localhost
R2_ENDPOINT	S3 Compatible Storage Endpoint	(Required)
JWT_SECRET	Secret for authentication tokens	(Required)

AI Decision Flow

cogniCV uses a chunked processing model to evaluate candidates efficiently while staying within AI rate limits.

1. Ingestion: Resumes are uploaded and parsed into structured JSON via Gemini.
2. Trigger: When a recruiter starts "AI Screening", the server fetches all candidate snapshots and the job description.
3. Chunking: Candidates are split into groups (default: 10) to optimize prompt performance.
4. Prompt Engineering: A "Master Screening Prompt" is sent to Gemini, containing the Job Criteria and Candidate Batch.
5. Scoring: Gemini returns a JSON array with:
   • matchScore: Weighted aggregate.
   • subScores: Skills (40%), Experience (30%), Education (15%), Relevance (15%).
   • reasoning: Human-readable justification.
6. Incremental Updates: The server saves results after each chunk, allowing the UI to show a real-time progress bar.
7. Final Ranking: Once all chunks are processed, candidates are ranked globally by score.

Assumptions and Limitations

Assumptions

• Resume Language: The system assumes resumes are in English for optimal Gemini performance.
• Parsing Status: Candidates must be successfully "parsed" before they can be "screened".
• Infrastructure: Assumes RabbitMQ and MongoDB are reachable as per the .env configuration.

Limitations

• Rate Limits: Heavy batches (hundreds of resumes) might hit Gemini's RPM limits; a 1-second delay is implemented between chunks to mitigate this.
• File Types: Currently supports .pdf for resume parsing and .csv for batch ingestion.
• Chunk Size: The system is tuned to process 10 candidates per AI call to balance speed and evaluation depth.
• PDF Quality: Text extraction quality depends on the underlying PDF structure (scanned images without OCR may fail).

© 2026 cogniCV Team. All rights reserved.