AI/ML Engineer with 2+ years of hands-on experience contributing to production-grade AI systems, RAG pipelines, and intelligent automation solutions. I have worked on designing, developing, and deploying ML solutions using transformer-based models, with a strong focus on reliable MLOps practices.
Building intelligent systems that solve real-world problems
🎓 B.Tech in Computer Science & Engineering
Swami Vivekanand Institute of Engineering & Technology • CGPA: 8.5/10
📍 Banur, Punjab, India
Building AI solutions at scale for leading organizations
Developed and maintained PHP-based web applications with an emphasis on backend performance, clean architecture, and reliable API integrations. Worked on responsive UI integration and optimized server-side logic to improve application efficiency and user experience.
Production-grade AI solutions across multiple domains
Built an end-to-end deep learning pipeline for classifying product images as Mobile Phones or Laptops using TensorFlow and MobileNetV2. Implements complete ML workflow including dataset preparation, training, evaluation, and deployment-ready inference via FastAPI. Features binary classification with transfer learning, achieving high accuracy on custom datasets for e-commerce automation.
Developed a machine learning system to detect deepfake/synthetic audio using Wav2Vec2 embeddings and classical ML classifiers. Achieved 92.86% accuracy with Logistic Regression on the Real vs Fake Human Voice dataset (70k samples). Pipeline extracts 768-dimensional feature vectors, handles variable-length audio, and implements preprocessing with StandardScaler normalization. Trained and compared Logistic Regression (best), SVM, and Random Forest models.
Architected an enterprise-grade multi-agent RAG system using LangGraph orchestration. Implements specialized agents for document retrieval, synthesis, fact-checking, and response generation. Features dynamic routing, agent collaboration, and context-aware memory management with 95%+ answer accuracy on domain-specific queries.
Fine-tuned BERT and T5 transformer models for domain-specific sentiment analysis. Implemented transfer learning, data augmentation, and advanced preprocessing. Achieved 94% F1-score on custom dataset with balanced precision-recall. Deployed with FastAPI for real-time inference.
Fine-tuned LLaMA 3.1 8B for classifying Linux commands and natural language into predefined intents. Built for AI terminal assistants and DevOps automation. Achieved 96% accuracy using LoRA fine-tuning with custom prompt-completion dataset.
Built a production-grade podcast processing system with speaker diarization, multi-host/guest identification, and automatic music filtering. Implemented real-time line-level editing, WebSocket-based progress tracking, and chunked long-form processing using a sliding-window approach for LLM limits. Integrated LLM-driven summarization, sentiment analysis with timestamps, and semantic search via Pinecone, with transcripts securely stored in AWS S3.
Built an intelligent complaint handling system using CrewAI multi-agent framework. Automatically processes text and audio inputs, classifies issues, verifies against policies, generates responses, and integrates with CRM. Reduced response time by 70% while maintaining quality.
Engineered a sophisticated n8n workflow with dual AI agents, persistent MongoDB memory, and intelligent routing. Implements context-aware conversations, webhook triggers, and modular architecture for scalable automation across multiple domains and use cases.
Developed a remote control system for Hisense TV and Fire TV using ADB and MQTT protocols. Enables seamless device communication, Android automation, and real-time command execution for smart home integration.
Built an ML-powered web scraping system to handle dynamic popups, CAPTCHA detection, and anti-bot measures. Trained a CNN-based popup detection model enabling automated interaction and seamless scraping across 50+ websites, achieving 98% accuracy using Selenium with headless Chrome.
Built a high-performance speech-to-speech system with industry-leading latency. Uses Deepgram for STT, Groq LLM for rapid inference, and ElevenLabs for natural TTS. Implements streaming responses, ChromaDB for knowledge retrieval, and optimized pipeline achieving consistent sub-2-second response times.
Developed a production-ready chatbot with LangChain and ChromaDB. Features voice interaction, real-time streaming via WebSocket, conversation memory, and automatic HTML transcript generation sent via SMTP. Handles context across sessions with personalized responses.
Built a production-grade video conferencing app with FastAPI and WebRTC. Features instant meeting creation, multi-participant support, text chat, user authentication, OTP recovery, and optional recordings. Optimized for low latency and high concurrent user capacity.
Comprehensive skill set spanning AI/ML, automation, scraping, and backend development
Open to exciting opportunities and collaborations
+91 7783805286
+91 9576243008