Skip to content

Latest commit

 

History

1 Commit

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

Hi, I'm Ricky 👋

I lead AI adoption at a listed company — in practice that means I'm the whole team: strategy, build, rollout, and the awkward conversations with the security reviewers. A fair amount of it I'm still figuring out as I go.

What I keep coming back to is the agent harness: the layer around the model rather than the model itself. Scheduling, memory, skills, evaluation, guardrails.

My working hypothesis, still being tested, is that model capability stopped being the bottleneck a while ago — and that what holds adoption back inside a company isn't the model not being smart enough, but the layer around it not being reliable enough yet.

What I'm trying to figure out

  • Getting agents into environments that hold secrets. Sandbox boundaries, permission matrices, approval interrupts, and the paths that quietly escalate privilege.
  • Making token cost predictable. Tiered routing, context compression, and reasoning budget allocated by task complexity.
  • Letting legacy systems absorb AI. Twenty years of process and data don't get a rewrite. The agent has to meet them where they already are.
  • Making outcomes measurable. Without evals there's nothing to iterate against, and as far as I can tell that's where most enterprise AI efforts quietly stall.

Open source

I contribute mostly around guard boundaries and data integrity — less "add a feature", more "where can this be bypassed, where does it hang, where does it drop data without saying so". It's slow work that doesn't demo well, but it's the part I find most interesting, and I learn a lot from the reviews.

Area
Command execution guards privilege and wrapper prefixes, path handling, option parsing
Checkpoint & rollback making "restored" actually mean restored
Sandbox & network policy proxy interference, honest network posture reporting
Gateway access control isolation and addressing across profiles
Runtime resilience idle timeouts, session state reclamation, health checks
Config & CLI lossless parsing, consistent key validation
Web UI & i18n rendering, form state, integration tests, locale catalogs

OpenSquilla — token-efficient microkernel agent. A local router sends each turn to the cheapest model that is good enough, which maps directly onto the constraint I keep running into at work. → my pull requests

Hermes Agent — self-improving agent with a built-in learning loop; it generates skills out of its own usage and keeps iterating on itself. → my pull requests

Stack

Python TypeScript Claude MCP Ollama Vue Next.js PostgreSQL Supabase Docker Cloudflare pytest

Also reach for React, Tailwind, Swift, SQLite, pandas, NumPy, Vercel, Stripe, Power BI and Streamlit depending on what the problem needs.

Say hi

Open an issue or a discussion. Always happy to compare notes on harness design, evals, or getting an agent past a security review — I learn as much from other people's answers as from my own.

About

个人主页 README

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors