Summary
I build AI systems end to end: multi-agent orchestration, MCP servers, model-forensics tooling that measures what an LLM actually computes, and the React front-ends that put them in front of people. I orchestrate dozens of coding agents at once, which is how one person ships and maintains this much. Before that I was the only engineer at a $3 billion company: every system was mine, and I was the one people escalated to.
Skills
Independent Engineering
Software I design, build, and operate in production, 2020 to present. github.com/asuramaya.
- osiris. Coordination and memory infrastructure for a fleet of 20+ AI coding agents. An event-sourced Postgres ledger where every fact records how it was obtained, exposed over MCP, with hybrid retrieval over full-text and embedding similarity.
- heinrich, chronohorn, decepticons. A model-forensics instrument measuring what an LLM actually computes, for evaluation and safety research; the multi-GPU experiment platform under it; and the PyPI-published kernel of sequence-model primitives that grew out of both.
- handlingtheloop. A serverless browser DJ app mixing public YouTube tracks. React and TypeScript front to back. Reverse-engineered a controller's undocumented MIDI protocol, then traced a silent hosting-provider IP block dropping the app's own fetches.
- hector-vector. A browser-based SVG editor, live and free at hector-vector.com. The desktop build adds an AI raster-to-vector pipeline on Real-ESRGAN, plus background removal and offline fonts.
- monsterhouse. A real-time livestream clip generator shipped as a SaaS. One focal pipeline follows the hottest stream while cheap monitors trigger focus swaps. Eight ML models share one 8GB GPU.
- lilguy, neo, kast. Shipped tools. lilguy is a FastAPI and Cloudflare search engine indexing 8,000+ postings; neo audits what agents like Claude Code leave on a machine; kast unifies AirPlay, Miracast, and Chromecast behind one Linux tile.
Experience
The entire technology function for a construction holding company managing $3B+ in active real estate development across 7 operating subsidiaries and 3 offices. No other engineers.
- Wrote the integrations between business systems. Worked with vendor engineers to learn each platform's API and data contracts, then built against them. Automated shop operations against the CollectiveData API. Built ingestion and extraction against Deltek ERP specs, tying field budgeting and operations to accounting in the office.
- Built a PowerShell and Python automation suite. It ran user and data lifecycle management end-to-end and auto-triaged most of a 20-ticket-a-day queue. Monitoring scripts surfaced visibility gaps across the device fleet.
- Owned identity, security, endpoints, and networking across 1,000+ endpoints for all seven companies. Migrated legacy infrastructure to Entra ID, Intune, and Defender, landing each rollout in three to six months.
- Ran the vendor stack and the security program. Directed partner engineers at Cisco, Netskope, Abnormal Security, Arctic Wolf, and Deltek. Contracted Arctic Wolf for monitoring and routine red-team exercises. Served as primary technical liaison to executives.
Endpoint support and configuration for a national homebuilder running ~90 offices across roughly 20 U.S. states.
- Handled Windows, Microsoft 365, and Zendesk ticketing for a high-volume user base across a distributed national footprint.
Temporary pandemic-response assignment supporting COVID-19 vaccination sites operated by Harris County, the third-largest county in the United States.
- Kept check-in, registration, and data ingestion systems running through peak surge, supporting high-volume public operations under sustained time pressure.
Cover Letter
Loading…
Sincerely,
Hector Chavez
Application Answers
Loading…