AI Systems, LLMOps & Agentic Workflows

Designing, evaluating, and monitoring LLM-powered systems for reliable knowledge work.

Focus

I build and evaluate LLM-powered systems that combine retrieval, orchestration, multimodal understanding, automated evaluation, and monitoring. The goal is not just to call a model API, but to design systems that can be tested, improved, and trusted in real workflows.

Core Capabilities

• Design and deployment of LLM-powered AI systems, including RAG, agents, and multi-agent orchestration.
• Multimodal ML systems for text, image, and embedding-based understanding.
• Hallucination detection, drift monitoring, automated evaluation, and regression testing with tools such as LangSmith and Langfuse.
• Agent architectures with LangChain, LangGraph, CrewAI, and Microsoft AutoGen.
• LLMOps and model lifecycle management: evaluation, monitoring, versioning, prompt management, and release discipline.
• Vector databases and retrieval systems, including FAISS, Pinecone, and LlamaIndex.
• LLM fine-tuning workflows, including PEFT/LoRA and QLoRA.

How I Present This Work

AI systems work should be demonstrated through artifacts:

• Architecture diagrams.
• Evaluation datasets and test suites.
• Retrieval quality metrics.
• Failure mode analysis.
• Monitoring plans.
• Reproducible notebooks or small deployable demos.
• Clear tradeoffs around cost, latency, quality, and risk.

Evidence On This Site

• RAG evaluation system
• Agentic research assistant
• Multimodal retrieval system