Orchestrated five parallel agent frameworks (Strands, Pydantic AI, Microsoft Agent Framework, Agno, Mastra) with a Google ADK orchestrator to build a small site end-to-end.

A QA agent opens each build in a real Chrome browser via Playwright MCP, verifies it works, and triggers one bounded auto-fix round for anything broken.

Four autonomous trading agents, each backed by a researcher agent, that read market data, use tools via MCP, and place trades on a schedule.

Built with the OpenAI Agents SDK and multiple MCP servers for accounts, market data, web search, and memory; a FastAPI backend and Vite dashboard visualize live portfolios and traces.

A multi-agent research assistant: a Clarification Agent scopes a vague question into a goal, then a Research Manager orchestrates planner, search, writer, and evaluator agents as tools.

The evaluator checks the final report against the original goal and triggers a bounded rewrite loop before emailing the finished markdown report.

A coordinated team of specialist agents builds a personalized interview briefing site from a company name, job posting, and interviewer info — grounded only in facts from your resume.

A research agent gathers public sources, then topic agents write company, interviewer, STAR-story, and mock-interview pages in parallel, with retries and guardrails against inventing experience.

A multi-agent tool that checks a job posting honestly against a CV, only drafting outreach when the fit is real, then has a second LLM review the email like a recruiter before it goes out.

Also drafts a LinkedIn note and tracks every company, contact, and status in SQLite, with a safety guard that defaults all sends to your own inbox.

Built a discrete-event simulation to analyze visitor flows and operational bottlenecks across a two-day music festival.

Analyzed service times and visitor behavior, fitted statistical distributions to observed data, and evaluated operational alternatives using KPI-based analysis.

The model was used to identify queue bottlenecks and compare operational scenarios to support data-driven resource-allocation decisions.

Built a linear regression model to explain Spotify song popularity from audio features, through exploratory analysis, preprocessing, and formal model diagnostics.

Performed variable selection, dummy and interaction variables, AIC-based model selection, and assumption testing (normality, linearity), improving fit with targeted transformations.