Independent Architect · Bulldog Technology Consultants
Agentic platforms, retrieval pipelines, and the secure cloud infrastructure underneath them — taken from prototype to production. Four decades of building, three CTO tenures, and a current book of work that lives on the frontier of applied AI.
What I do
I work as architect, lead, or individual contributor — and I'm equally comfortable explaining the tradeoffs to investors and a board as I am writing the code that proves them out.
Production agentic systems on agent SDKs like Claude Agent SDK and OpenCode — in-process MCP tool servers, subagent orchestration, extended thinking, parallel tool execution, and per-agent cost guardrails. Workflow orchestration with LangGraph and LangChain.
Retrieval-augmented generation over PostgreSQL + pgvector (HNSW), FAISS, and Chroma. Hybrid BM25 + vector retrieval with Reciprocal Rank Fusion, Neo4j knowledge graphs, and long-lived agent memory with importance-weighted decay.
Multi-account AWS as Infrastructure-as-Code: Bedrock, Lambda, API Gateway, RDS, S3, Secrets Manager, VPC. Private data access via SSM bastions — no public IPs, no static keys — with IAM least-privilege and compliance-isolated accounts.
Branch-guarded deploy automation for multi-account production stacks, ECR-versioned container builds, and pytest suites with separated unit/integration tiers — plus the evaluation infrastructure that proves an LLM system actually works in production, with tracing and cost instrumentation in Langfuse.
RESTful and streaming APIs in FastAPI + asyncpg, Django REST Framework, and NestJS. SQL design and performance optimization on PostgreSQL and MySQL, plus Salesforce integration over the REST API and private warehouse tunnels.
Control and UI software for FDA-track medical devices, including real-time monitoring at high sample rates. Native iOS and Android across two decades — from Objective-C and Swift to Kotlin and cross-platform stacks.
Selected work
Recent contracts run through Bulldog Technology Consultants. Earlier roles span medical, gaming, mobile, and the early commercial internet.
An end-to-end AI pipeline for a leading booker of private-jet travel that automates the operator request-for-quote workflow — outbound RFQs and inbound reply parsing alike.
Ranks hundreds of charter operators for fit with per-operator rationale, then composes and sends personalized RFQ emails. On the inbound side it matches operator replies to the right trip via multi-stage hybrid retrieval over 900+ candidate trips, classifies them, and extracts structured proposals. A Neo4j context graph feeds the LLM stages through MCP tool calls. Cut LLM spend on the extraction path by ~80% with a mail-boilerplate sanitizer, verified live in Langfuse.
Production AI systems for advisory clients — including an agentic knowledge-graph and retrieval platform built on the Claude Agent SDK.
An in-process MCP tool server with a streaming (SSE) FastAPI + asyncpg backend, hybrid BM25 + vector retrieval over PostgreSQL/pgvector and Neo4j; an LLM-driven scanner that orchestrates subagents across Gmail and databases to surface AI-automatable business processes as a knowledge graph; AI-powered legal document analysis; a Trino-federated bridge over multi-tenant warehouses; and a real-time voice agent prototype.
Control and user-interface software for a pole-mounted system that delivers therapeutic medicine directly to the brain.
Continuously monitors pressure and flow through the patient's cerebrospinal space at high sample rates, recording and analyzing the data in real time. Used successfully in multiple animal trials nationwide; the prototype has been delivered to a contract manufacturer for FDA approval.
Scalable, secure web services for storing and analyzing results from a portable electronic COVID-19 test.
Built for HIPAA and GDPR compliance under FDA regulation. The device — Alveo's “be.well” test — won the XPRIZE Rapid COVID Testing Competition. Also co-designed a reversible infusion pump and UI for precise movement of cerebrospinal fluid, validated in animal trials.
Co-founder, CTO, and VP of Engineering of one of the defining products of the early commercial internet.
Built the first prototypes with a small team, then grew and led 85 engineers and testers shipping high-performance internet servers, clients, and information feeds. Helped close over $50M in private investment and grew the company to $18M in annual revenue and 1.5M subscribers. Spoke on panels at Stanford and industry conferences.
A secure, scalable backend game engine hosting Bingo, slots, and casino-style games with real cash payouts.
Built in Python / Django / DRF with an extensive unit-test suite and AWS deployment automation. Live and serving clients, integrated with a wide range of third-party services — PayPal, Stripe, Prizeout, Sentry, New Relic, geolocation, and more.
Two decades more — Stratasys, Goodpath, Drync, Sonde Health, AARP, GrabCAD, and others — are in the full CV.
Track record
I've been Chief Technology Officer at three companies — PointCast, AdelaVoice, and GorillaPark — and a co-founder at two of them. I've built and led engineering teams from a handful to 85, and presented technology to investors, boards, partners, and the press.
Since 2012 I've run Bulldog Technology Consultants, delivering software and technology services to startups and large companies alike — often bringing in and managing trusted subcontractors when an engagement needs more than one pair of hands.
My career runs from Pascal and Ada compilers at Digital Equipment Corporation and wafer-scale manufacturing control, through the early internet, to today's frontier AI work. The throughline: shipping systems that are secure, scalable, and real.
Based in Cambridge, Massachusetts. I work remotely with clients anywhere, communicate clearly with technical and non-technical audiences alike, and speak business/conversational French.
Let's talk
I take on a small number of contract engagements at a time — architecture, build, or technical leadership. If you're working on something hard, I'd like to hear about it.
