AI / Agent Engineer

Multi-agent research systems

• I designed a lead agent that planned research and coordinated specialist agents. Each specialist returned structured evidence for the final answer.
• I built pause and resume around persisted execution stages, plans, and task state. Recovery checks guarded against duplicate tool calls and charges. Persisted events and SSE let clients replay results after reconnecting.
• I added tool authorization, session concurrency limits, run budgets, and output filtering. Tool execution checked agent and user permissions, available capabilities, and task relevance. Retrieved evidence carried its source information.

Python · asyncio · Letta · FastAPI · SSE

AI products, mobile delivery, and infrastructure

• I built an AI memoir workflow covering interviews, memory, fact-checking, and image generation. I separated ASR, TTS, and LLM providers from the business logic, then connected real-time voice, background generation, and PDF export to Expo clients for iOS and Android.
• I integrated computer-vision pipelines and isolated inference environments behind FastAPI services. The work included stereo-camera analysis and real-time video processing using MediaMTX and FFmpeg. I also built a native Swift client for visual-analysis reports.
• I set up local GPU services, Gitea, Docker registries, CI/CD, and OpenTelemetry with the LGTM stack. I trained nontechnical colleagues to use AI tools in their daily work.

LangChain · FastAPI · PostgreSQL · Redis · Celery · Expo · Swift · Docker

Music tools and content automation

• I packaged music-generation models into local choir-teaching tools. The workflow turned scores and reference audio into voice-part accompaniment and demonstration audio, with Core ML inference on a Mac Studio.
• I built n8n and LLM workflows for topic collection, outlines, platform-specific rewrites, review, and scheduled publishing. Shared API nodes handled authentication, retries, failure alerts, and logs.

Python · Core ML · n8n · LLM APIs