|
Location: Seattle, WA
Remote: Yes
Willing to relocate: Yes (SF, NYC for the right team)
Technologies: Go, Python, Rust, TypeScript, Kotlin | LLM agents, multi-agent orchestration, MCP, RAG, evals, GPTQ/quantization, Triton | Postgres, Redis, Kafka, gRPC, Kubernetes | AWS, GCP, Cloudflare | Cadence/Temporal, Playwright/CDP
Résumé: https://elninja.com/resume
LinkedIn: https://www.linkedin.com/in/elninja
GitHub: https://github.com/ibarrajo
Email: alex [at] elninja.com
I build agent systems. A decade in production distributed systems (Uber, DoorDash, Jobscan); the last year spent designing and shipping autonomous, LLM-driven infrastructure. Looking for a hands-on senior IC role on an AI-infra / agent-platform / applied-AI team.Background: at Uber, worked on Cadence (open-source workflow engine, 12B+ workflows/month internally) — executed multi-region failovers on domains doing billions of ops/day, shipped non-determinism detection via shadow replay, rebuilt cadenceworkflow.io, onboarded 50+ teams. At DoorDash, refund rules engine for McDonald's/Chipotle (500K refunds/month). At Jobscan, Interim Head of Eng: 97% → 99.99% uptime through an AWS migration, $475K/yr recovered via payment A/B testing. I've operated durable-execution systems at scale and have opinions about where that paradigm earns its keep and where autonomous agents are the better tool — which is what I build now. Shipped this past year: - Maquina: a structured "cognitive language" + control plane for coordinating heterogeneous LLM agents. Full EBNF grammar, an evaluation harness, and a runtime that treats multi-agent coordination as a first-class protocol instead of glue code. - Meridian: 29k-line TypeScript autonomous goal-graph executor (Postgres + MCP) that decomposes high-level charters into plan nodes and runs LLM-backed agents on a live dashboard, with budget circuit-breakers and governance enforcement. - Pursuit (co-founder, AI Scalathon Seattle 2026): agent-to-agent recruiting — a 3-min structured interview producing ranked multi-dimensional match scores. Ran live: 51 reports + 120 real matches against Microsoft, JPMorgan, Uber, CoreWeave, Adobe, Block, Boeing, Lululemon, SpaceX. - ApplyPilot (OSS, github.com/ibarrajo/ApplyPilot): multi-provider LLM orchestration with cost-aware routing across Gemini/OpenAI/Anthropic; Claude Code + Playwright automation across 40+ ATS workflows. - OpenAI Parameter Golf: 14+ PRs to the 16MB / 10-min / 8×H100 LLM-training challenge (best valid submission val_bpb 1.1354). Hands-on with Int5/6 GPTQ, score-first test-time training, Triton kernels, and Hessian-based calibration. I want a team where the hard problems are agents, evals, and inference — and shipping daily is the norm. |