2
1
India
1 year of experience
Abhishek Rajput | AI & ML Enthusiast Passionate AI/ML engineer with a Minor in AI from IIT Ropar and hands-on experience in Deep Learning, Computer Vision, and Model Deployment. Built high-accuracy models, including CNN-based image classifiers, fraud detection systems, and medical AI solutions. Skilled in Python, TensorFlow, PyTorch, and MLOps (FastAPI, Docker, MLflow). Excited to solve real-world challenges with AI and collaborate on innovative, impact-driven projects! 📧 [email protected]

CaptionAI is an intelligent multi-agent video captioning platform that transforms any video into engaging, creative captions across multiple styles. Built on LangGraph's stateful agent orchestration framework, the system deploys a team of specialized AI agents working in parallel to deliver fast, high-quality results. How It Works: Video Ingestion – Users submit single or multiple video URLs through a bold neo-brutalist React interface built with Vite and Tailwind CSS. Agentic Pipeline – A LangGraph-powered agent graph orchestrates the workflow: Validator Agent checks URL validity and video duration (30–120 seconds). Downloader Agent fetches videos from cloud storage to local temp storage. Audio Extractor Agent uses FFmpeg to extract audio tracks (gracefully handles silent videos). Frame Extractor Agent samples keyframes every 5 seconds using FFmpeg. Transcription Agent converts speech to text via OpenAI Whisper API. Vision Analyzer Agent uses Gemini 2.5 Flash to describe visual scenes in each keyframe. Content Merger Agent combines transcript and visual descriptions into unified context. Caption Generators run in parallel using Groq's LLaMA 3.3 70B model to produce four distinct styles simultaneously: Formal (professional, grammatically correct) Sarcastic (witty, clever mockery) Humorous Tech (programming/IT humor) Humorous Non-Tech (general audience humor) Quality Checker Agent validates caption length and relevance. Results – Users receive beautifully styled caption cards with copy-to-clipboard functionality. Technical Architecture: Backend: Node.js + Express.js with LangGraph for agent orchestration, BullMQ for job queuing, FFmpeg for video processing, and Zod for request validation. AI Models: Groq (text generation), Gemini 2.5 Flash (vision), OpenAI Whisper (transcription). Frontend: React + Vite with Tailwind CSS. Deployment: Vercel serverless functions for backend API, Vercel static hosting for frontend.
13 Jul 2026