Platform AI Architect · Principal Engineer

Jordan
Carson

Platform AI Architect and Principal Engineer with 13+ years delivering production AI systems at scale. Built DevGPT, a coding agent harness serving 6,500+ engineers at J.P. Morgan Chase, architected as a clean separation between the reasoning harness ("brain") and an isolated remote sandbox controller ("hands") for secure, multi-tenant agentic code execution. Engineered prompt caching to 90%+ hit rates and a distributed telemetry pipeline handling 50K events/min.

View DevGPT ↗ Read Blog →
Impact
6,500+
Engineers on Platform
Engineering, product & senior management
90%+
Prompt cache hit rate
Anthropic Claude · DevGPT
32M
Peak cache reads / min
Against a 30M TPM budget
50K
Telemetry events / min
OTel streaming gateway · DevGPT
Technical Skills

Hover for details

Career
2025–Present
Current
Lead Architect & AI Systems Engineer — AI4Tech
J.P. Morgan Chase
  • Lead architect for DevGPT's core agentic systems: the coding harness (reasoning, tool use, memory) and a separate remote sandbox controller for isolated code execution
  • Coding agent harness — custom agentic loop, tool-calling, and multi-step execution comparable in scope to Claude Code, fully decoupled from execution
  • Remote sandbox controller — pool-managed, session-aware Kubernetes pods on EKS with MCP Bridge and SOCKS5/socat network isolation; the harness never holds credentials or network access directly
  • DevGPT Cloud — the collaboration layer: co-ideate with an agent alongside teammates, push/pull sessions, observe live sandbox execution, browse a team agent marketplace
  • DevGPT powers Technology Lifecycle Management (TLM) — firm-wide agentic maintenance across Java codebases today, expanding to Python, Golang, TypeScript, and Terraform
  • Sole AI4Tech engineer on the AWM AI Center of Excellence, defining agentic development standards across the firm
  • Designing AWM Agentic Training (in dev) — a Trailhead-style platform using AI as a Socratic brainstorming partner and grader
  • Temporal-backed durable agent orchestration
  • Four-layer memory system (Org / Team / User / Session) — in development
  • 2-week AWS org migration with 30-minute planned outage — zero data loss
2021–2025
Senior Applied AI Engineer — Asset & Wealth Management
J.P. Morgan Chase
  • Built DevGPT from the ground up within AWM's Architecture & Investment Technology team
  • ~90% prompt cache hit rates on Anthropic models; 32M peak cache reads/min against a 30M TPM budget
  • Two independent, mission-critical LLM provider services — Anthropic via AWS Bedrock, OpenAI via Azure OpenAI — no shared prompt router; provider selection is a deliberate, user-facing decision
  • Custom telemetry gateway: producer → SQS → consumer → staging → processors, feeding a federated service (Aurora writes, real-time attribution) and a separate analytics service (read replica, reporting) — ~50K events/min
  • Kubernetes API gateway on EKS using Cilium with per-LOB auth, rate limiting, load shedding
  • 65% AWS cost reduction — Karpenter/EKS node group migration + smart prompt engineering & context caching
  • Knowledge Graphs & Event-Driven News Analytics Platform — entity resolution at ~900ms/article
  • Sm@rt RFP — weeks to minutes for client proposal generation
2015–2021
Data Scientist & Analytics Developer
BNP Paribas · Global Markets
  • Lead Americas data scientist
  • Salesperson 360 dashboard adopted org-wide
  • Contributor to pyGCA internal ML library
2013–2015
Software Engineer / Analyst
UBS Securities · Group Data Services
  • Client Onboarding, KYC, & Data Warehousing
  • Chief Economics Office intern — cubic spline interpolation for defined benefit pension plan valuation