About
I build LLM systems that people can actually trust in production.I am an AI Engineer with 3 years of experience shipping production Python, now focused on the layer most GenAI projects skip: the evaluation and guardrails that decide whether a model's output is safe to act on. I am currently deepening this through IIT Kharagpur's Executive PG in Generative AI and Agentic AI.Recently I placed top 5 of 9,000+ participants at a national 24-hour AI hackathon presented by Cursor and Deutsche Telekom Digital Labs, building a decision-automation platform that kept AI on a fallback path behind a deterministic rule engine, with every decision returned as a full audit trail.What I do:• LLM Application Development: end-to-end apps with OpenAI and Anthropic APIs, FastAPI backends, and React front ends• Agentic Workflows: multi-agent systems on LangGraph with schema-validated tool registries and policy checkers gating every action• RAG & Vector Search: Pinecone, FAISS, and MongoDB Atlas Vector Search with chunking, re-ranking, and semantic search• Evaluation & Guardrails: ablation studies, adversarial test suites, failure taxonomies, and hermetic tests that run without network or API keys• Production Python: FastAPI, Django, Docker, Kubernetes, and cloud AI on AWS Bedrock and SageMakerAlongside the AI work, I spent three years as a Software Research Engineer building real-time systems for UAVs in C++ and Python, so I know what it takes to ship performance-critical software that has to work the first time.What I care about most is AI that is reliable, explainable, and accountable, not just impressive in a demo.If you are hiring for AI Engineer, LLM, or Applied GenAI roles, let's talk.
n8n experience
n8n CloudAI agents in n8n
ONNXCUDAGraphics Processing UnitAmazon Web Services (AWS)PyTorchFastAPIGenerative AIPrompt EngineeringRetrieval-Augmented Generation (RAG)Large Language Models (LLM)QtData PipelinesResearch ComputingDockerResearch SoftwareContinuous Integration and Continuous Delivery (CI/CD)Engineering ResearchScientific ProgrammingResearch EvaluationC++Python (Programming Language)Pandas (Software)NumPyProject ManagementSQLOpenAILangChainHugging FaceAnthropicOllamaReplicateAWS BedrockGitHubPostgreSQLMongoDBPinecone
Workflow showcases
Reviewed work samples with client-sensitive information removed.
ExpertPolicy-Guarded Multi-Agent System with Guardrail Ablation
A multi-agent system built to make autonomous agent actions safe and measurable. Every tool call passes through a schema-validated tool registry and a policy checker that gates each write, so the agent cannot take an unauthorized or malformed action. Failure handling is the core of it: an adversarial suite probes for injection and unsafe writes, the policy layer blocks them, and a fallback path keeps the agent from dead-ending when a step fails. I ran an ablation across six guardrail configurations to isolate which controls actually move reliability, measuring pass rate, policy violations, steps per task and cost per task, so the safety claims rest on numbers, not assertions.
6-variant guardrail ablation measured pass rate, violations, steps and cost per task; the policy checker and adversarial suite drove unauthorized writes to zero in tested runs.PythonLangGraphmulti-agent orchestrationschema-validated tool registrypolicy engine
