GenAI orchestration with frontier models
Built Python integrations with LangChain and provider libraries that connect to frontier models from Anthropic and OpenAI, and used LangGraph libraries to orchestrate multi-step GenAI workflows that turn manual steps into automated ones. These models are used in every phase, from research and development through testing and rollout.
- LangChain
- LangGraph
- Python
- Workflow automation
RAG prototyping on local models
Used a locally hosted LLM on Ollama to develop and test retrieval-augmented generation approaches in Python. Loaded documents into a vector database, then experimented with chunking, embeddings, retrieval, and prompt templates to see what produced reliable answers.
Event-driven ingestion modernization
Standardized Kafka topics and schemas, implemented dead-letter and retry patterns, and built Spring Boot microservices that validate and transform feed payloads, with better observability and throttling across thousands of daily workflows.
- Kafka
- Spring Boot
- Microservices
BPMN orchestration at scale
Authored Camunda 8 BPMN models and Spring Job Workers coordinating multi-tenant ETL pipelines, with clean hand-off to Kafka and REST services, idempotency, and status outputs.
Agentic development workflows
Our web platforms, mobile games, and this site are built with coding agents such as Claude Code and Codex, orchestrated through project-level AGENTS.md and CLAUDE.md instructions that encode architecture, conventions, and guardrails. Agents plan, implement, run builds, and verify their own changes in a real browser before review, and independent work runs in parallel sessions.
- Claude Code
- Codex
- AGENTS.md
- Parallel sessions
Scheduled AI automations
Recurring agent runs that operate the BrainFunHub business: a daily article generator that checks the full content history for overlap, verifies facts, and inserts unpublished drafts into Supabase; daily social publishing to Pinterest, Facebook, and X with live-post verification and a posting log; and weekly SEO reports built from Google Search Console and Bing Webmaster Tools.
- Scheduled agents
- Supabase
- SEO
- Browser automation
Reusable agent skills
A library of packaged skills that turn multi-step work into repeatable pipelines: a printable PDF maker with puzzle validation and rendered-page inspection, a store-listing generator that reads a finished PDF and writes ecommerce copy and SEO metadata, an article writer, and a Stripe product creator that previews every change and requires explicit confirmation before going live.
- Agent skills
- Python
- Stripe
- Content pipelines
Human-in-the-loop guardrails
Automation designed to be safe by default: secrets loaded only into the process that needs them, preview-then-confirm for anything that touches money or production, drafts instead of publishing where judgment matters, idempotent inserts, and agents that stop and report on logins, CAPTCHAs, or unfamiliar UI instead of guessing.
- Safety
- Idempotency
- Secrets
- Review gates