Aram Barnett

AI Product Engineer — Staff / Founding · Agent Pipelines, Evals & Entity Systems

San Francisco, CA

aram.barnett@gmail.comLinkedInGitHubBarnett Labs

Summary

Staff-level product engineer with 10+ years shipping AI-native products, fintech/crypto systems, and developer platforms from 0 to production — hands-on in code every day (TypeScript, Python). I build complete agent systems: LLM pipelines with eval harnesses, guardrails, memory, and retrieval, plus entity graphs with scoring across four production domains (wallets, tokens, markets, creators). Recent receipts: an autonomous pipeline turning raw scripts into validated budgets and schedules in production; a 17,000-entity market-intelligence graph over 22 automated connectors; a monetized multiplayer iMessage AI game built and shipped in 72 hours.

Core Skills

AI & agent systems:Multi-agent orchestration; LLM pipelines (Gemini, Claude, DeepSeek); eval harnesses for output accuracy; guardrails (compliance/rate floors); agent memory across tool calls; retrieval (web + domain data); entity extraction & NER; knowledge graphs; MCP servers; prompt engineering; AI cost instrumentation
Engineering:TypeScript, Python, Node.js, React/Next.js/Vite, React Native, Firebase, PostgreSQL, Prisma, Redis, WebSockets/SSE, REST/GraphQL
Infrastructure:Google Cloud (Cloud Run, Pub/Sub), AWS, Docker, Kubernetes, Fly.io, Vercel, CI/CD, observability; systems at 100K+ req/min and 99.8% uptime
Payments & fintech:Stripe usage-based billing, Apple Pay, AML/attribution systems, market data, Solidity/wallets

Experience

CurbilyCTO & Lead Product Engineer

Jul 2024 – Present

AI operating system for vertical-drama content production — script breakdowns to crew sourcing.

  • Architected and personally built the multi-agent pipeline (Gemini) turning unstructured scripts into validated budgets, schedules, cast/shot lists, and scene breakdowns — evals check outputs against test sets for budget and shot-list accuracy, guardrails enforce union-rate floors so generated payments never fall below union minimums, retrieval spans web/local search and SAG rate data, and memory persists context between tool calls.
  • Built the enrichment engine producing 2,900+ creator profiles from raw agency rosters — photos, stats, scoring, media kits — refreshed nightly at ~$40/run.
  • Cut cloud cost ~80% while improving processing speed ~90%; instrumented AI spend, attributed 95% of it to one unwatched workload, and rebuilt pricing to a 97% gross margin.
  • Designed an internal agent organization — 15 runbook-driven agent roles across engineering, QA, design review, content, and ops — letting a 3-person team ship like 20, with overnight runs and 6am status reports.
  • Own end-to-end delivery daily: architecture, API design, prompt engineering, data modeling, UI/UX, Stripe usage-based billing, deployment, and code review.

WrldBuilderFounder & CEO / Lead Product Engineer

2022 – 2024
  • Venture-backed Web3 product studio ($1M raised, managed to plan); built blockchain-integrated 3D applications (Unity, Solidity, Web3.js); scaled a global contractor marketplace to 160+ developers in under 30 days.

Nervos NetworkSenior Product Manager, Special Projects

2021 – 2022
  • Designed and shipped a low-gas NFT standard adopted across the developer ecosystem; led 20+ engineers to ship a dynamic-NFT Web3 game; built cross-chain proxy APIs and developer tooling.

PowerboardCo-Founder & Technical Lead

2020 – 2022
  • Play-to-earn gaming infrastructure; backend handled 100K+ requests/minute at 99.8% uptime; grew Discord-bot downloads +60% MoM to 40% retention; fielded multiple acquisition offers.

TRM LabsCrypto Intelligence Lead

Dec 2019 – May 2020
  • Built data-collection tooling for AML/blockchain-attribution systems; gamified analyst workflows, cutting cost ~15% and raising collection rate ~50%; ran a global team of 15 with KPI dashboards (MySQL, GCP, Airflow).

Human VenturesEntrepreneur in Residence

2018 – 2019
  • Built 0→1 ventures at a NYC startup studio (mobile collectible game + 4D-hologram companion hardware); designed a permissioned Hyperledger Fabric network across 70+ universities; led technical diligence for the studio's blockchain investment arm.

CoinScoreFounder / Head of Product

2016 – 2018
  • Token-scoring models across 1,100+ crypto assets (hundreds of data points per asset); grew 0 → 10,000 MAU; led acquisition and partnership conversations with major financial companies.

Selected Projects — shipped end to end, solo

  • Slop Fight & Sprouts — iMessage-native AI games (2026): Built, shipped, and monetized a multiplayer battle game in 72 hours — LLM judge scoring, generative image renders on players' real selfies, video finishers, and $2.99 Apple Pay purchases in-chat; strangers paying within 48 hours of launch (Cloud Run, slopfight.lol). Forked the chassis into a second game in one evening; Sprouts development is grant-funded by the Aptos Foundation.
  • MarketMotion — prediction-market intelligence (2025): 17,000+ entity knowledge graph over 22 automated connectors; AI entity discovery/NER and hop-proximity linking (DeepSeek-V3) with confidence scoring; cross-venue arbitrage across Polymarket (CLOB) and Kalshi; real-time alert engine and B2B API (Redis caching, leader election, batch processing).
  • Barnett Labs (2026 – ): Fixed-price AI engagements delivered as working software — AI cost/margin audits, entity engines, agentic-inbox installs. labs.arambarnett.com

Leadership, Affiliations & Education

Launched one of the first US Bitcoin investment funds (2012) · Advisor to the SEC on token-sale regulation · Founding Member, Chamber of Digital Commerce Token Alliance · Guest Lecturer, Morgan State University · Advisor, Blockchain @ University of Michigan

Morehouse College — Studies in Economics & Physics (GPA 3.6)