AI Demos

A joke generator, travel suggestions, a website summarizer and an LLM Arena. Uses Groq (GPT OSS 120B) and OpenAI (GPT-4o mini), with a Flask web page.

Chains, memory and agents: a website summarizer, a chat that remembers, and a shop agent. Built with LangChain and OpenAI (GPT-4o mini), with a Flask web page.

The full RAG pipeline: embeddings, chunking, vector search, reranking and chat with a PDF. Built with LangChain, Chroma and GPT-4o mini, with a Flask web page.

Live demos of three Docker projects, from one ML container to an agent system that runs many services. Built with Flask and Docker Compose.

Agents that use tools and reason in several steps. Built with the AWS Strands Agents SDK on Amazon Bedrock (Converse API), with a Flask web page.

An AI coach for mock interviews. A ReAct agent calls tools to check each answer for STAR structure, relevance and filler words as you go. It remembers the session and gives tips during it. Uses OpenAI.

Sorts shipment problems by urgency. It applies a compensation policy, decides who to escalate to, drafts messages, and keeps a live daily ledger with KPI totals.

Trade-offs between prompting styles, and how models give in to pressure (sycophancy). Also The Refund Bench: a 4-stage agent pipeline that settles disputes, where 3 judges vote.

Measures what makes AI output reliable: forcing a schema, chain-of-thought, routing by tool schema, and whether a model admits a failed tool call or makes up data.