All Projects
Agentic AI systems across five patterns: ReAct, Plan-and-Execute, Multi-Agent, Self-Correcting, and Tool-Use. The in-site showcases run interactively on this domain; the Python prototypes are research notebooks.
Recursive Portfolio Chatbot
A self-correcting RAG agent that answers questions about my background using a real LangGraph correction loop — pgvector retrieval (with mock fallback), Claude Sonnet generation, and a 'See Thoughts' toggle exposing the full trace.
Deep Research Agent
A plan-and-execute autonomous research agent — decomposes a topic, searches sources, cross-references findings, and emits a downloadable PDF. Every node streams its thoughts over SSE so the live trace is the demo.
Software Architect Crew
A 3-agent crew (PM → Coder → Reviewer) that takes a product prompt to reviewed code. Mermaid.js flowchart shows which agent currently holds the token; each agent's reasoning streams live over SSE.
Agent Playground
Live ReAct tool-use playground — Claude Haiku plans each turn against a small tool registry (weather, local time, portfolio recommender) and streams every thought, action, and observation over SSE.
Agent Skills Dashboard
Sample observability UI for agent runs — KPI strip, model latency-vs-success scatter, and per-tool accuracy. Metrics are placeholders today; the page is the aggregation surface that would sit over a production telemetry table.
Customer Service Agent
Prototype Python/Jupyter retail agent for sunglasses inventory and transactions — formal tool registry, argument canonicalisation, DuckDB over pandas, assertive validation, and propose-vs-commit transaction tooling driven by GPT-4o plan reflection.
Market Research Agentic Team
End-to-end AI marketing pipeline in Python/Jupyter — multi-agent market research, internal product matching, AI image generation, copywriting, and an executive markdown report. Includes a reflection loop that monitors execution and dynamically revises the remaining plan.