Product engineering
Fintech and healthcare in São Paulo: order processing cut from 300 to 5 seconds at FCamara, 80% fewer bug reports at Provi. Taught 200+ students web development with Meta's Hack Station.
STAFF SOFTWARE ENGINEER · WBD · INFLUENCER · LUXURY PRESENCE
Staff engineer at Warner Bros. Discovery (streaming CDNs, GraphQL at 15–25M requests a day), Influencer (creator discovery, API latency down 97%) and Luxury Presence. 54 merged PRs across 16 open source repositories. Now: native ZCode chat and a Hermes Agent harness in agent-orchestrator, both in review.
WARNER BROS. DISCOVERY · INFLUENCER · LUXURY PRESENCE · 54 PRs IN 16 REPOS
Eight years of production engineering. The scale changed every year — the standard never did.
TRACK RECORD
HOW I SHIP
Scale: multi-region, multi-cloud CDNs at Warner Bros. Discovery — AWS, Fastly, GCP, Akamai — serving hundreds of millions toward a 99.99999% availability target.
Speed: creator discovery at Influencer — AI inference with creator, audience and demographic filters, vector search over embeddings, third-party API latency down from 30s to 900ms.
Rigor: test infrastructure that holds — Testcontainers against live PostgreSQL, SQS and Redis, 100+ test cases, an SDK replacing 10,000+ duplicated lines across six microservices.
Agents: native ZCode chat and a Hermes Agent harness for agent-orchestrator, both in review upstream — shipped the same way I've always shipped: reviewed, tested, merged.
SELECTED WORK
ALSO: trigger.dev (ESLint plugin + CLI tests) · Qdrant read-only API keys · zio-json 256-bit integers · Permit.io docs upgrade · gbrain · oss-site
NOW
ZCode chat integration (#6003) and Hermes Agent harness with ACP chat (#6635) in review on OrchestratorInc/agent-orchestrator.
gbrain in daily use.
oss-site catalog at 77 projects / 8 groups.
AS OF OCTOBER 2026
TIMELINE
Fintech and healthcare in São Paulo: order processing cut from 300 to 5 seconds at FCamara, 80% fewer bug reports at Provi. Taught 200+ students web development with Meta's Hack Station.
Senior engineer on the streaming platform: legacy CMS replaced (saving $5M a year), GraphQL APIs at 15–25M requests a day, Kafka and SQS pipelines for HBO, TNT and DC, 100+ CVEs remediated across 50+ repositories.
Cloud/SRE on the multi-region CDNs behind hbomax.com, cnn.com and discoveryplus.com (95M viewers on HBO alone); Luxury Presence: Testcontainers, 100+ test cases, an SDK replacing 10,000+ duplicated lines.
Staff engineer on creator discovery — AI inference, vector search, third-party API latency from 30s to 900ms. Now contributing agent-orchestrator upstream.
FAQ
No. Agents are tooling, not a substitute for judgment. Every change still goes through the review and tests I've used since shipping to hundreds of millions of users — a diff merges only when it's right.
Running many agent sessions as one system: spawn, supervise, steer, review, recover. I build the harness itself — my ZCode chat integration and Hermes Agent harness are in review on agent-orchestrator.
Yes — it's the resume, in public form: 54 merged PRs across 16 repositories, including rustfmt, screenpipe, cal.diy and Qdrant. Full table on my GitHub profile.
The scale I know: products with real users and real consequences — streaming, fintech, creator platforms — that want AI-native speed without giving up review rigor.
Email, one click from every page. Issues and PRs on my repos reach me too.
Partly — but the claims on it are all resume: WBD, Influencer, Luxury Presence, and public PRs you can open. No build step, static by design, works from file://.