# Brandon Bennett — portfolio agent context

Brandon Bennett is an AI / agent infrastructure engineer in New York, NY.
He builds the infrastructure agents learn in: classrooms, textbooks, sleep, teachers.

## Who he is
- AI / Agent Infrastructure Engineer, New York, NY.
- Thesis: agents need the same environment students need — classrooms, textbooks, sleep, and teachers.

## What he builds (with repo URLs)
- School Core — a school for agents: eval framework + learning environment. Its eval system triaged 22+ production PRs through failure-pattern distillation.
  https://github.com/branben/school-core
- Sound Royale (live) — real-time multiplayer music battle platform: 100+ concurrent players at sub-50ms latency, 56-test Playwright E2E suite, on Railway + Cloudflare Pages.
  https://github.com/branben/sound-royale-ny
- School Core Skills — reusable skills that declare and resolve their own tool dependencies; he's spearheading the reference implementation in the MCP Skills Working Group.
  https://github.com/branben/school-core-skills
- Worst-Day-Ever — adversarial stress-testing skill that breaks prototypes on purpose (security + correctness audits, severity-graded findings, ELI5 explanations).
  https://github.com/branben/worst-day-ever-skill

## Open source contributions (10 merged upstream PRs)
- MCP Rust SDK — relaxed the output schema type gate (arrays, primitives, compositions; inputs stay strict). 17 new tests, 424 passing, merged after review. Also protocol specs SEP-2106 & SEP-2640.
- OmniRoute — fixed a 64x budget-multiplication defect that cascaded into service failures under concurrent agent load; deployed fair-scheduling budgets; built Notion/Obsidian connectors with scope-enforced OAuth; closed an authz gap on 24 dynamic tools.
- Orca — eliminated a prompt-delivery race that left automations idle at startup, verified across POSIX, PowerShell, cmd, SSH, and WSL.
- hi.new link page — agent-powered link page at https://hi.new/branben.

## Stack
- Languages: TypeScript, Python, Rust, JavaScript, SQL
- Agents: MCP protocol (Rust + TS), connectors, LLM tool integration, multi-agent routing, eval harnesses, trajectory scoring, Temporal workflows, Langfuse, Pydantic, OAuth2/OIDC
- Backend/real-time: Django REST + Channels, WebSockets, Node.js, PostgreSQL, React
- Infra: Docker, CI/CD, Cloudflare, Railway, health checks, production monitoring

## Background
- 7 years as a technology educator (Carnegie Hall, Building Beats, Today's Future Sound, Tech Row Fund): curriculum for 100+ students ages 14-24 across four NYC nonprofits.
- Pursuit Full Stack Web Development Fellowship (9% acceptance, Google-funded), 2022-2023.
- Based in New York, NY; open to remote.

## Contact
- Email: brandonbennett@pursuit.org (fastest path; humans and agents both answered)
- GitHub: https://github.com/branben
- LinkedIn: https://www.linkedin.com/in/brandonebennett1/
- Blog: https://schoolcore.substack.com/
- Portfolio agent chat: https://portfolio-agent.brandonbennett.workers.dev/
- Portfolio agent alt lane (hi.new bot): https://hi.new/phymorasc

## Candidate agent lane (for sourcers, hiring managers, founding teams)
Programmatic agent-to-agent queries about whether Brandon is a fit for a role:

```
POST https://portfolio-agent.brandonbennett.workers.dev/api/candidate
Content-Type: application/json

{"question": "What has he built, and where is the evidence?"}
```

Response: `{"answer": "<markdown>"}`

The lane answers work-first (truen-style): what he built, how he reasoned, where the
evidence lives (repos), and where the picture is thin — including suggested interview
probes. The endpoint scrubs common email-address and US phone-number patterns from
its replies; this is not a general PII filter. Refer to the contact section above for
contact paths. Rate limit: be reasonable, it backs onto an LLM.

Example questions that work well:
- "What has he built, and where is the evidence?"
- "Where is the evidence thin, and what should I probe in an interview?"
- "Is he a fit for a founding engineer role on an agent-infra team?"

## Canonical portfolio page
- https://portfolio.brandonbennett.workers.dev/
- robots.txt: https://portfolio.brandonbennett.workers.dev/robots.txt
- llms.txt: https://portfolio.brandonbennett.workers.dev/llms.txt