Senior Software Engineer and AI Engineer based in Lima, Peru. I build AI agents and LLM-powered automation on top of solid backend engineering, and I take them to production.
- 6+ years building backend systems, including a platform with 500K+ users and 30M+ daily transactions.
- Designed an LLM automation pipeline that cut incident triage time by 92%, from 20 minutes to under 90 seconds.
- Currently pursuing an MS in Applied AI at Golden Gate University.
My main project. An autonomous factory that designs, builds, tests, audits and documents production-grade software overnight, while a human approves what gets built and reviews every pull request.
The models write and correct the code; a deterministic pipeline decides what is accepted, under a hard budget, with every token and dollar recorded.
- 330 projects delivered as pull requests across four assembly lines.
- 84% delivered with no human fix (308 of 368 started items). Every hand fix is recorded as an intervention and excluded from that rate.
- Model-agnostic: the same pipeline and gates run on Anthropic, Google Gemini, Vertex AI and DeepSeek, measured side by side. API cost starts at USD 0.12 per project.
- Production quality gates on every project: lint, types, tests with at least 80% coverage, documentation standard, secret scan, dependency audit, an independent specification audit, and CI on Linux and Windows.
- A team of roles, not one prompt: analyst, architect, generator, fixer, auditor, docs writer, security engineer and visual QA, plus deterministic planning, gates and publishing.
See how it works and the measured results: live page · showcase repo
The factory itself (source code, prompts, rules and operating data) is private. What you see here is what it has built.
Where one generation is not enough, the factory's architect splits the work into modules with frozen interfaces and builds them in dependency order.
| Project | What it is |
|---|---|
| u401-lsm-kv-engine | LSM-tree key-value storage engine |
| u402-linearizability-checker | Checker for linearizability of concurrent histories |
| u403-wasm-interpreter | WebAssembly interpreter |
| u404-cdcl-sat-solver | CDCL SAT solver |
| u405-crdt-json | CRDT for collaborative JSON documents |
| u406-linear-regex | Linear-time regular expression engine |
| u407-formula-engine | Spreadsheet formula engine |
| u408-datalog-engine | Datalog engine |
| u409-columnar-sql-engine | Columnar SQL engine |
| u410-git-objects | Git object store |
Unreal Engine agent toolsets for the Unreal Engine 5.8 MCP server, delivered with no human fix: asset intake, captures, render profiles and performance probes.
All delivered projects: browse the catalog
- Languages: Java, Go, Python, TypeScript, Rust
- AI and agents: Anthropic, Gemini and Vertex AI, DeepSeek, OpenAI, LangGraph, CrewAI, RAG, tool calling, structured outputs, evals, MCP
- Backend and architecture: Spring Boot, FastAPI, Node.js, microservices, Hexagonal Architecture, DDD, CQRS
- Data and cloud: PostgreSQL, pgvector, Redis, AWS, Google Cloud, Azure, Docker, Kubernetes, GitHub Actions
- AI-assisted development: Claude Code, Cursor, GitHub Copilot
Open to remote roles and B2B contracts.


