About Me
Senior LLM Engineer with 5+ years of hands-on experience building, fine-tuning, and shipping production-grade LLMs and agentic AI systems. Core expertise spans the full LLM lifecycle — pre-training data pipelines, SFT/RLHF post-training, RAG architecture, multi-agent orchestration, and systematic prompt engineering.
Skills & Technologies
Work Experience
LLM Pre/Post-Training Specialist & AI Evaluator (Freelance)
Curated and structured large-scale pre-training corpora for language models, maintaining 98%+ annotation accuracy across 30k+ examples. Delivered SFT datasets and preference pairs for LLM fine-tuning projects. Evaluated Hindi-language AI responses, achieving a 100% Job Success Score.
LLM Pre/Post-Training Specialist & AI Evaluator (Freelance)
Curated and structured large-scale pre-training corpora for language models, maintaining 98%+ annotation accuracy across 30k+ examples. Delivered SFT datasets and preference pairs for LLM fine-tuning projects. Evaluated Hindi-language AI responses, achieving a 100% Job Success Score.
Senior AI Data Trainer & Annotator
Annotated over 50,000 samples with 98%+ accuracy and led a sub-team of 6 annotators. Developed RLHF workflows for conversational AI and performed various NLP tasks including sentiment analysis and multi-turn prompt evaluation.
Projects
Agentic RAG System with LangGraph + RAGFlow
Built a stateful multi-agent RAG system using LangGraph for agent orchestration and RAGFlow for document ingestion and retrieval, achieving sub-400ms E2E latency and <4% hallucination rate.
LLM Pre-Training Data Pipeline & SFT Fine-Tuning
Designed an end-to-end LLM pre-training data pipeline for a 7B-param domain LLM, followed by SFT using LoRA on SageMaker, reducing perplexity by 18% and hallucination rate by 42%.
Multi-Agent LLM Orchestration Platform
Engineered a multi-agent framework using LangGraph stateful graphs and persistent Redis memory, achieving a 55% reduction in task completion time.
Computer Vision Defect Detection Pipeline (Deep Learning)
Trained and deployed a YOLOv8-based defect detection model for industrial quality control, achieving 94.3% [email protected] and deployed as a real-time FastAPI microservice.
Automated LLM Evaluation & CI/CD Quality Gate
Created a real-time evaluation dashboard tracking hallucination rate and answer relevancy, integrated with GitHub Actions CI/CD to cut regression detection time from days to minutes.
Decibench — Voice Agent Benchmarking Framework
Developed an open-source benchmarking framework for voice AI agents, supporting automated multi-turn conversation simulation and latency profiling.
NearestDoctor — AI-Powered Healthcare Search
Built a full-stack healthcare discovery platform with LLM-powered symptom-to-specialist matching and real-time doctor availability, achieving 99.5% uptime.
Education
B.Tech in Computer Science & Engineering
Computer Science