Jaiden Sy

AI Workflows · Consulting

AI that gets past the pilot

I build the AI workflows companies actually run.

Most AI projects stall in the same place. The demo works, everyone's impressed, and then nobody will let it near the real systems, the real data, or a customer. That's the part I do. I build the workflow end to end, wire it into what you already run, and put the permissions, audit trail, and blast-radius controls around it that make someone comfortable turning it on. By day I wire services into a top-5 US bank's auth and API-gateway stack (OAuth 2.0, Apigee, PingFederate); after hours I built and open-sourced an MCP gateway that governs how agents touch real tools.

Start a conversation See the work

01 How I can help

How I can help

Two shapes, mostly. A process your team still runs by hand, or an AI feature that works in the demo and can't get cleared for real use. Both end up in the same place: something that runs on its own, with agent authentication, tool authorization, delegated access, and MCP-gateway governance around it so it stays that way.

01 · Build it · End to end

Build the workflow

Design and build a production AI workflow from empty repo to launch: agents that do real work, tool-calling, MCP, retrieval, and the integrations into the systems you already run, with the auth, tests, and observability it needs to survive real users.

02 · Harden it · Production-safe

Make it not break

You have a workflow that works in the happy path. I make it safe for production: tool authorization, rate limits and cost controls, audit trails, graceful failure, and the edge cases nobody tested.

03 · Advise · Retainer

A second senior set of eyes

Ongoing architecture reviews and on-call sanity for your AI infrastructure, so the thing you shipped keeps working, and the next thing gets designed right the first time.

02 Selected work

Selected work

Systems I designed and built end to end: the same work I do for clients. Arbiter is a live MCP gateway; Engram runs at home.

OBJ-01 · Arbiter ● Live

A self-hosted MCP gateway: one governed door between AI agents and every tool they call.

Agents authenticate through Arbiter instead of holding raw credentials: deny-at-call-time RBAC with a full audit trail on every proxied call, an encrypted secrets vault that checks authorization before it ever decrypts a key, and a Redis semantic cache that cut repeat tool-call latency from ~5.7s to ~33ms. FastAPI and PostgreSQL behind a React and TypeScript front, on Railway and Vercel. Designed, built, and launched solo.

  • fastapi
  • react / typescript
  • postgresql
  • redis
  • railway
  • vercel
arbiterai.dev →
Arbiter architecture diagram AI agents such as Claude Desktop, Cursor, and custom agents route through Arbiter, which applies auth, quotas, audit, and caching, then reaches tools including GitHub MCP, Postgres MCP, and internal tools. claude desktop cursor custom agent ARBITER auth · quotas audit · cache github mcp postgres mcp internal tools

every tool call authenticated, metered, and logged

OBJ-02 · Engram ○ Complete

An agent-orchestration daemon that runs real work from a plain text message.

Routes a natural-language request to the right coding agent, runs it headless, and replies with the result: model-tier routing, task classification so non-code work skips the code path, and watchdogs that keep the loop alive unattended. The same orchestration and reliability patterns that keep production agent systems from falling over.

  • python
  • agent orchestration
  • llm routing
  • launchd

03 The day job

Four years, one trail.

A top-5 US bank · San Francisco · in progress. Enterprise authentication, identity, and API-gateway integration. The unglamorous discipline I carry into the AI work.

  1. Act I · 01/2023 – 02/2025 · Software Engineer

    • Built cross-language auth and logging libraries, standard across 100+ engineering teams.
    • Built a GitOps config and secrets-sync service (HashiCorp Vault); config-only redeploys, adopted by 10+ teams.
    • Reusable Python templates cutting new-service setup ~20 hrs each.
    • Inner-source across 8 teams, ~30 PRs merged.

  2. 02/2025 · Promotion · Software Engineer → Senior

  3. Act II · 02/2025 – present · Senior Software Engineer

    • Wire services and micro-frontends into the org's API gateway (Apigee) and OAuth provider (PingFederate) using OAuth 2.0 PKCE and Authorization Code flows: the identity and gateway layer other teams' apps authenticate through.
    • Migrated ~9 services PCF → OpenShift/K8s in ~2 weeks; wrote the migration playbook now used by 50+ engineers.
    • Own the micro-frontend templates nearly every team in the org builds on; enforce org-wide version gates.
    • Shipped metrics, attestation, and audit micro-frontends tracking framework usage org-wide.
    • Coordinate ~25 production releases a year.

04 Skills

The instruments.

Six areas. Solid dots are daily drivers; the hollow ones are still forming.

Backend

Python, FastAPI, Django 4, REST APIs

OAuth2 / OIDC, Flask

Go

Frontend

React, TypeScript, Micro-frontends

JavaScript, WebSockets

Data

PostgreSQL, SQL

Redis, MongoDB, RabbitMQ

Cloud · Ops

Docker, Kubernetes / OpenShift, CI/CD

Azure (AZ-900), Railway, Splunk

Terraform, K8s operators

Security

RBAC / multi-tenancy, Secrets vault (Vault), Auth libraries, Audit logging

OAuth2 / OIDC / SSO

AI Infra

Agent orchestration, MCP gateways, Agent identity / tool authz, Audit logging

Semantic caching, LLM agent tooling

LangGraph, pgvector

DAILY DRIVERS: Python · FastAPI · React · TypeScript · PostgreSQL · Docker

05 Contact

Start a conversation.

Have an AI workflow to build, or one that's stuck in pilot and can't get cleared to go live? That's the conversation I want. Email is fastest.

hello@jaidensy.com