Antonio B. De Castro⌗

Spokane, WA · abdecastro@protonmail.com · linkedin.com/in/antonio-b-decastro

Summary⌗

Full-stack engineer with 12+ years of product experience across hardware, logistics, media, and AI. Spent the last four years at Humane building on-device AI systems for the AI Pin. Now focused on LLM application architecture, agentic workflows, and AI product consulting for early-stage companies and enterprise teams.

Experience⌗

Independent AI/LLM Consultant⌗

Self-Employed · Feb 2025 – Present · Remote

Working with startups and product teams on LLM integration, RAG pipelines, and AI product architecture. Engagements span greenfield app development, evaluation frameworks, and helping engineering teams scope AI features realistically.

  • Designed a multi-agent document processing pipeline for a legal tech client using LangGraph and Claude 3.5, reducing manual review time by ~60%
  • Built a RAG-based internal knowledge tool for a 200-person SaaS company; reduced Tier 1 support volume by 30% in first 90 days
  • Advising two seed-stage founders on AI product strategy and LLM vendor selection

Senior AI Engineer⌗

Humane · San Francisco, CA · Jan 2021 – Feb 2025

Joined as employee #28 to build the software layer for the AI Pin — a screenless wearable running a persistent AI assistant. Worked across inference infrastructure, multimodal input processing, and the Cosmos AI platform that powered the device.

  • Led development of the on-device intent classification system that routed queries between local inference and cloud LLM endpoints based on latency and connectivity constraints
  • Built the voice pipeline abstraction layer handling ASR → intent → response → TTS across 3 underlying model providers with <300ms target latency
  • Owned the Laser Ink Display rendering pipeline integration — the projected UI system that sat on top of Palm OS
  • Contributed to the AI context persistence system that maintained session state across the stateless HTTP interactions between Pin hardware and Cosmos backend
  • Worked closely with the ML team on fine-tuning evaluation for the assistant’s domain-specific response patterns (calendar, contacts, communications)
  • Spent the last 18 months post-launch on triage and system stability; shipping bug fixes into a product with universally poor market reception is its own kind of engineering education

Lead Software Engineer⌗

Katerra · Menlo Park, CA · Aug 2020 – Jan 2021

Katerra was building vertically integrated construction — factories, supply chain, and software under one roof. Joined to work on internal tooling connecting factory output to project management. The company filed Chapter 11 five months after I left.

  • Built internal APIs connecting factory production schedules to project management tooling used by site supervisors across 40+ active job sites
  • Led a small team (3 engineers) integrating third-party BIM data into Katerra’s procurement platform to automate material quantity takeoffs
  • Rewrote a brittle Python ETL pipeline handling supplier pricing data; reduced daily failure rate from ~15% to near zero

Senior Software Engineer⌗

Quibi · Los Angeles, CA · Jan 2020 – Aug 2020

Eight-month stint during the launch window. Quibi shipped in April 2020, two weeks into COVID lockdowns. Worked on the cross-platform playback infrastructure for the HQ (Horizontal/Vertical) format switching system.

  • Built client-side logic for seamless video orientation switching — rotating the phone mid-playback triggered a different cut of the same scene
  • Contributed to the CDN edge caching strategy for dual-format content delivery; assets were stored and served as independent streams, not dynamically transcoded
  • Watched $1.75B in VC evaporate from a front-row seat

Staff Software Engineer⌗

Zume Pizza · Mountain View, CA · Mar 2018 – Jan 2020

Zume was putting robotic arms in pizza delivery trucks. I was brought in to own the fleet software layer connecting kitchen automation, truck routing, and order management.

  • Designed and built the fleet telemetry API that aggregated real-time truck location, oven temperature, and order state into a unified ops dashboard used by kitchen and dispatch teams
  • Built the demand prediction pipeline (Python/scikit-learn) that fed pre-bake scheduling; the model reduced wasted pies during peak hours by ~22% over a three-month test window
  • Led a ground-up rewrite of the driver app (React Native) after the original vendor delivered something unusable; shipped v2 in 11 weeks
  • Managed two junior engineers; both are now senior ICs at their current companies

Senior Software Engineer⌗

Shyp · San Francisco, CA · Jun 2015 – Feb 2018

Shyp picked up your packages and handled shipping for you. I joined when the company was operating in 4 cities and building fast. Owned the courier-side mobile infrastructure and parts of the routing backend.

  • Built the courier app (iOS, Swift) from scratch to replace a third-party tool that couldn’t handle variable-duration pickups — adoption across the courier fleet was complete within 6 weeks of launch
  • Redesigned the job dispatch algorithm to account for package volume estimates; reduced instances of couriers arriving with insufficient vehicle space by 34%
  • Shipped real-time package tracking for customers; previously updates were batch-processed, which made the “on-demand” positioning hard to defend
  • Part of the team during the New York launch and the subsequent contraction — I’ve now been through two startup wind-downs and have opinions about the warning signs

Software Developer⌗

Juicero · San Francisco, CA · Sep 2013 – May 2015

Employee #12. Built the web platform and connected device dashboard for a $699 WiFi-enabled cold-press juicer. Yes, that Juicero.

  • Built the customer-facing ordering portal (Rails/React) for subscription juice pack delivery, including a fulfillment integration with the third-party warehouse
  • Developed the device registration and telemetry backend that tracked press activity, firmware versions, and pack usage across the connected device fleet
  • Contributed to the iOS app that let users schedule presses remotely and receive low-pack notifications
  • The press was eventually shown to be squeezable by hand. I knew this before Bloomberg did.

Quality Assurance Engineering Intern⌗

Adobe · San Jose, CA · Jun 2011 – Sep 2011

Summer internship on the Omniture (now Adobe Analytics) QA team. Wrote test plans and automated regression scripts for data collection and reporting features. First exposure to software at enterprise scale.

Education⌗

Cornell University · Ithaca, NY Bachelor of Science in Engineering (B.S.E.) · 2011 Concentration in Electrical and Computer Engineering

President, Cornell Electronic Circuits Society (ENG) · 2010–2011

Skills⌗

Languages: TypeScript, Python, Swift, Go, SQL AI/ML: LangChain, LangGraph, OpenAI API, Anthropic API, Hugging Face, RAG architecture, vector databases (Pinecone, Weaviate), fine-tuning evaluation, prompt engineering Web: React, Next.js, Node.js, Rails, REST/GraphQL API design Infrastructure: AWS (Lambda, ECS, RDS, S3, SQS), Docker, Terraform, CI/CD Data: PostgreSQL, DynamoDB, Redis, Kafka, dbt