Building production AI systems at scale — pipelines, agents, and observability that ship.
I'm a Generative AI Engineer at Tata Consultancy Services, building and shipping AI systems used by millions of users. My work lives at the intersection of agentic pipelines, LLM safety, and production observability — not just prototypes, but systems that run in the real world.
- Multi-agent orchestration using LangGraph and LangChain — designing agent-to-agent communication, tool selection, and execution planning
- LLMOps & observability — instrumenting 35+ LLM/ML metrics via OpenTelemetry, Jaeger, MLflow, LangKit, and Evidently AI for drift detection and real-time tracing
- LLM safety & guardrails — modular pipelines for PII masking, toxicity detection, bias filtering, and prompt injection defense across enterprise deployments
- RAG at scale — hierarchical retrieval systems with 100,000+ clustered embeddings, tuned for hallucination resistance under <3s latency
- MCP integration — building tool-augmented agents that connect LLMs to real-world APIs and data sources
| Project | What it is | Stack |
|---|---|---|
| 🔍 Multi-agent research assistant | LangGraph agents for search, summarization, and fact-checking | LangGraph · FastAPI · Streamlit |
| 📊 LLMOps monitoring dashboard | Open-source observability layer for LLM apps | OpenTelemetry · MLflow · Evidently |
| 🛡️ Modular AI guardrails | Pip-installable safety middleware for LLM pipelines | Guardrails · LangChain · HuggingFace |
| 🤖 Agentic code reviewer | MCP-powered agent that reviews GitHub repos automatically | MCP · LangChain · GitHub API |
Automated Grievance Redressal System @ TCS — AI pipeline serving 3.5M+ users, combining two-layer ML classification with hierarchical RAG over 100,000+ embeddings. Delivers hallucination-resistant answers in 2–3 seconds.
LLM Guardrails & Content Safety @ TCS — Modular safety filters deployed across 10+ enterprise AI use cases, covering PII masking, toxicity, bias, and gibberish detection. 100% data protection record.
LLM Observability Platform @ TCS — Instrumented 35+ LLM/ML metrics across distributed systems using OpenTelemetry, Jaeger, and MLflow with real-time drift detection.
RoadGuard (Telangana Govt. initiative) — YOLOv8 road analysis system identifying 3,000+ potholes across Telangana, contributing to a 25% reduction in road accidents in targeted areas.
AI / Agentic
LangChain LangGraph CrewAI MCP Ollama Azure OpenAI HuggingFace Transformers Guardrails
Observability / MLOps
MLflow Evidently AI OpenTelemetry Jaeger LangKit Loguru
ML / DL
PyTorch TensorFlow Scikit-Learn YOLOv8 MMAction2
Backend / Data
FastAPI PostgreSQL MongoDB Streamlit
Core
Python SQL Bash Linux Git
🏆 TCS AI Fridays 2025 — Winner, Gujarat Region · National Finals (Top 25 teams globally)
🥇 Amazon ML Challenge 2023 — Ranked 301st out of 4,898 teams nationally
📄 Published — Machine Learning: A New Perspective, presented at IEMPOWER 2022 · In publication at AIP Publishing
