Agentic workflows
I design tool-using agent loops with explicit state, safe actions, memory, logging, and review points.
MCP Playwright tool calling
~/ ai-integration-engineer
I build AI systems that turn messy tasks into working tools.
Agents, MCP tools, RAG pipelines, parsers, IDP workflows, and async backends - shipped as production-minded MVPs.
task
-> research
-> specification
-> agent.loop
tools: [mcp, rag, parser]
state: explicit
review: required
-> tests
-> deployable_mvp
output: report + working system 12+ public repositories MCP tools RAG over internal docs Local LLM workflows FastAPI / Docker / TypeScript
Capabilities
Systems around models, not just prompts.
The work is strongest where product ambiguity, AI constraints, and integration details meet.
RAG & knowledge systems
Search pipelines over internal documents: chunking, embeddings, BM25, hybrid retrieval, access-aware answers, and evaluation loops.
Qdrant FAISS BM25
Document AI / IDP
Long PDFs and unstructured documents become structured data through OCR, anchors, LLM orchestration, and async backends.
OCR PyMuPDF FastAPI
Business MVPs
Focused tools for domain workflows: imports, reports, dashboards, automation, and deployable prototypes.
TypeScript Docker PostgreSQL
Selected work
A curated casebook, not a repository dump.
Featured projects are chosen for AI integration signal: agents, MCP, RAG, IDP, local LLM workflows, and business MVP delivery.
Local LLM / Research Automation Public
research-local-xiaomi
Local research workflow for turning web evidence into structured reports with planning, critique, usage tracking, and reproducible outputs.
TypeScript CLI local LLM web search structured reports
Shows research pipeline design around local models and agentic review loops.
RAG / Full-stack AI Academic
neuro_apk
Full-stack AI employee prototype for agribusiness: RAG, local LLM integration, crawler, FastAPI backend, PostgreSQL, Qdrant, and Docker Compose.
Next.js FastAPI PostgreSQL Qdrant LM Studio Docker Compose
Demonstrates end-to-end AI system architecture from frontend to vector search and local inference.
AI Agents / Browser Automation / MCP Public
browser_agent_mvp
Local browser agent that controls visible Chromium through Playwright using ARIA snapshots, structured tool calls, memory, and logging.
Playwright MCP tool calling ARIA snapshots action logging
Demonstrates practical agent loop design for browser automation.
Business MVP / Data App Public
msfo-global-mvp
Business MVP for IFRS/MSFO-style financial workflows: data import, reports, charts, and export-oriented interface.
TypeScript frontend data import reporting charts
Shows ability to turn a domain workflow into a usable internal tool.
MCP / Enterprise RAG Client Work
Redmine MCP + VK Teams RAG
Private client work: MCP tools for Redmine and a role-aware VK Teams RAG bot over Wiki.js and PDF regulations.
Python MCP RAG VK Teams API Wiki.js PDF
12 Redmine tools.
How I build
Agentic development with review loops.
Most of my projects start as unclear tasks. I turn them into research notes, specs, implementation prompts, review loops, tests, and deployable MVPs.
- 01 Research
- 02 Specification
- 03 Agent implementation
- 04 Review
- 05 Testing
- 06 Deploy
Business automation
Where this helps
I help turn unclear operational tasks into small AI systems: internal assistants, document pipelines, RAG search, MCP tools, and fast prototypes that teams can actually test.
Discuss an automation ideaAutomate internal workflows
Replace repetitive manual steps with small tools, agents, and structured pipelines.
Prototype AI features fast
Go from idea to testable MVP without spending months on architecture theater.
Integrate LLMs into existing tools
Connect models to documents, APIs, browsers, task trackers, and business systems.
Currently building
Local research workflows, AI agents, this Astro casebook, and MCP/RAG experiments around practical business tasks.
contact
Bring a messy AI task.
I can help turn it into a scoped MVP, internal tool, research workflow, parser, RAG system, or agent integration.