AI Agent Systems Engineer β’ Tooling & MCP Architect β’ Automation Pipelines
Building deterministic interfaces, spatial grounding engines, and autonomous workflows for LLM agents.
I architect reliable, deterministic software systems bridging large language models and physical/digital documents:
- Agent Tooling & MCP Protocols: Designing zero-leakage, stdio-isolated Model Context Protocol (MCP) servers with strict JSON-RPC contracts.
- Document Geometry & Spatial Grounding: Eliminating LLM positioning hallucinations in complex forms and workbooks using hybrid vector extraction, Set-of-Marks (SoM) coordinates, and Virtual Excel grids.
- Zero-Pollution & In-Memory Computing: Guaranteeing pure RAM-based pipeline execution (
io.BytesIO, NumPy raster buffers) with zero intermediate disk clutter. - Automated Cloud Workflows: Serverless LLM curation engines and scheduled event processing (GitHub Actions, Telegram Bot APIs).
|
Pixel-Perfect PDF Form and Worksheet Filling Engine for AI Agents. Solves floating text, printed text collisions, and scanned workbook blindspots via Virtual Excel Matrix and Set-of-Marks snapping.
|
Serverless Cloud AI Digest & Tech Trend Curator. Scans GitHub trending repositories daily at 06:00 AM, filters low-signal repos using Google Gemini, and broadcasts clean technical briefings to Telegram.
|
- Languages & Core: Python (PyMuPDF, pdfplumber, NumPy, OpenCV, Asyncio), Modern TypeScript/JavaScript, C++, SQL.
- AI & Agent Infrastructure: Model Context Protocol (MCP), LLM Tool Integration (Claude Desktop, Cursor, Antigravity), Set-of-Marks (SoM) Grounding, Agentic TDD.
- Document Processing & Vision: PDF Vector AST Analysis, Morphological Raster Line Scanning, Baseline Snapping, Dynamic Font Kerning & Auto-Scaling.
- DevOps & Automation: GitHub Actions CI/CD, Serverless Architectures, Telegram Bot API, Docker.
- Deterministic Grounding over Heuristics: Give agents addressable coordinate grids (e.g., Virtual Excel Matrix
[S01]) rather than letting them guess floating-point bounding boxes. - Zero Disk Pollution: All intermediate raster buffers, transformations, and HTTP payloads are kept strictly in RAM (
io.BytesIO) β preserving disk hygiene and host security. - Surgical Precision: Minimal intervention surfaces with end-to-end TDD verification.