Brandon Garate
Applied AI engineer building agent platforms and evals. brandongarate177@gmail.com, (619) 361-9906.
Education
San Diego State University, San Diego, CA. B.S. Computer Science, Minor in Mathematics. Expected Dec 2027.
Skills
Agents and AI: agent harnesses (Claude Agent SDK, Claude Code), MCP servers and tool calling, evals and regression sets, multi-agent orchestration (Bedrock AgentCore), RAG pipelines, voice agents (LiveKit, ElevenLabs).
Languages: Python, TypeScript, Go, Swift, C++, SQL; FastAPI, FastMCP, React, Node.js, Django, tree-sitter.
Cloud: GCP (Cloud Run, Cloud SQL, Workflows, Firestore), AWS (Bedrock, Lambda, ECS), Terraform, Docker, Postgres.
Experience
Applied AI Engineer, part-time, Opdo, San Diego, CA. Sep 2026 to present.
- Built the agent platform that lets Opdo’s optical engineers hand lens simulations to AI agents and steer them from their own Claude chat. It’s an MCP server on Cloud Run with 31 tools over an append-only Postgres record, now in pilot on a live customer lens project.
- Wrote the worker that runs Claude Code headless through the Agent SDK against 5 pinned optics solvers. Hooks block the agent from editing the code that runs it, and a ledger logs every solve so each result traces back to its source.
- Designed a verifier agent that re-solves a job’s best designs on its own and grades the work against a checklist, and built its eval set of 11 regression cases, 6 with planted faults.
Member of Technical Staff, self-employed, VibeDrift, San Diego, CA. Jun 2026 to present.
- Co-building VibeDrift, a Claude Code plugin and MCP server that flags code that breaks a repo’s own patterns while the agent writes. Runs locally. 400+ paying users.
- Designed and led a blinded A/B experiment. Coding agents from 6 model vendors built the same ERP spec twice, 76 sprints and 60K lines per arm. The VibeDrift arm used 29% fewer output tokens (11.9M vs 15.4M), and the team published every result, nulls and one retraction included.
- Moved the scanner’s import graph, dependency and dead-code detection from regex to AST parsing, cutting false positives 70% (79 to 24), and extended detection to Go, Python and Rust.
Cloud Automation Engineer, internship, DishTV / EchoStar, Denver, CO. Jun 2026 to Aug 2026.
- Built a RAG pipeline indexing 27,000+ network documentation files into a Bedrock Knowledge Base, cutting vRouter troubleshooting from 15 minutes to under 30 seconds and turning per-device SSH work into fleet-wide batch runs.
- Architected a multi-agent system on AWS Bedrock AgentCore with 15+ tool-using agent workflows behind an MCP gateway with SigV4 auth and semantic tool discovery.
AI Engineer, internship, Ix-Infra, San Diego, CA. Apr 2026 to Jun 2026.
- Wrote tree-sitter-sas from scratch, then tested it on real SAS repositories and fixed two parsing gaps (82% and 78% miss rates) to reach zero parse failures across 32 tests.
- Shipped the SAS and R parsers for Ix’s ingestion pipeline, modeling macros and DATA/PROC steps as graph entities, which unblocked codebase indexing for biotech and biomed clients.
AI Engineer, contract, GradeFlow, San Diego, CA. Nov 2025 to Feb 2026.
- Owned an audio transcription pipeline in Go end to end, proxying client uploads to ElevenLabs voice agents and on to the downstream evaluation services. Used daily by 850+ students at 1.8s transcription latency.
Projects
CoTeX, SDSU ZIP Launchpad. Jul 2025 to Jan 2026. Built a codebase-aware Markdown editor on GitHub OAuth and webhooks through SDSU ZIP Launchpad, and ran 35+ customer interviews with working engineers, then shut it down when they showed it wasn’t sticking.
Scibble, scibble.app. Built an app that turns a whiteboard photo into a narrated math walkthrough, on an async video pipeline chaining GPT-4o Vision, OpenAI TTS and Manim rendering on AWS ECS Fargate with SQS, S3 and CloudFront.
Leadership
ColorStack @ SDSU, Executive Board. Led career-development workshops.
CTRL Hackathon, Organizing Committee. Ran a 120-attendee hackathon with industry sponsors and mentors.
