Agentic Frameworks
Multi-agent orchestration with LangGraph, LangChain, CrewAI, AutoGen. Tool-use, memory, planner/executor separation and MCP — decision-making systems shipped in production.
I ship generative AI, agentic frameworks, and automation pipelines. From LLMs to Diffusion, LangGraph to n8n — building AI production lines across industries.
Building an end-to-end agentic game-dev framework — delegating the full production line to agents: from idea and design doc, through style-consistent visuals and code, to audio and video/animation generation.
AI/ML Engineer // Data Scientist — I ship generative AI, agentic frameworks, and automation pipelines. From LLMs to Diffusion, LangGraph to n8n — building AI production lines across industries.
Multi-agent orchestration with LangGraph, LangChain, CrewAI, AutoGen. Tool-use, memory, planner/executor separation and MCP — decision-making systems shipped in production.
Always-on automation with n8n, Make, Zapier, and Temporal. AI positioned as decision-maker across email, CRM, and internal tool integrations.
Hybrid retrieval, rerank, graph-RAG, knowledge graphs. Neo4j, Qdrant, pgvector — enterprise answer quality grounded in your data.
ComfyUI, FLUX, SD, LoRA fine-tunes for brand-consistent visuals; Suno, ElevenLabs for audio; Runway, Kling, Sora for video. Game assets to ad creative.
End-to-end product vibe-coding with Claude Code, Cursor, Windsurf, v0, Aider. Custom MCP servers, bespoke agents, review automation, spec-driven delivery.
LLM + custom NMT (OpenNMT, Marian) hybrid pipelines. Style-guided, term-consistent, QA-able translation for games, e-commerce, docs.
LLM observability via Langfuse, LangSmith, Arize; evaluation via Ragas and DeepEval. Model comparisons, prompt versioning, cost/latency budgets.
Time-series and reinforcement learning for LTV, demand forecasting, supply-chain optimization. Production models for e-commerce and games.
End-to-end web game production and level-design automation. AI vibe-coding for rapid iteration, procedural levels, style-consistent asset generation.
Pit LLMs against each other on a chess board — tool-use, planner behavior, model comparison.
Before/after hair-transplant visuals with FLUX + LoRA. Control-weighted Diffusion pipeline.
GitHub Actions + LLM for automatic PR review. Spec-driven comments and diff-aware suggestions.
Computer vision experiment: pixel-level recoloring of beaded designs on canvas.