01
Multi-Agent System Build
Production agents that run your workflow end-to-end — without a human in the loop on every step.
AI Consultancy
Production AI agents that actually ship — and prove they work.
Multi-agent systems for founders and engineering leaders who've already been burned by demoware. Top 4% on the GAIA benchmark; deployed across energy, government, e-commerce, and logistics.
Built for energy operators, government agencies, and e-commerce teams — and the founders quietly hating their last AI vendor.
Services
Four ways I help teams get AI into production — and keep it there.
02
AI Reliability & Evaluation
Evaluation frameworks that prove your AI works before your customers find out it doesn't.
03
Architecture Review & Strategy
A senior read on your AI roadmap before you commit engineering budget to it.
04
AI Project Rescue
Audit and recover stalled AI initiatives — from drifting offshore teams to POCs that won't ship.
Selected Work
Production-grade systems, not POCs. Expand any row to read the case.
Top 4%
GAIA public leaderboard
Public Benchmark · Autonomous AI
Most AI agents stall the second they need to act without a human pressing a button. To prove I could close that gap, I built and ranked an autonomous research agent on GAIA — the leading public benchmark for end-to-end agent autonomy. The system solves 300 multi-step research tasks (web research, code execution, file analysis, verification) with zero human intervention, and placed in the top 4% globally — above public submissions from OpenAI, Google, NVIDIA, and Microsoft. The architecture pattern — orchestrator routing to specialist agents with a critic loop — is the same one I deploy in client production systems where reliability is non-negotiable.
- Role
- Lead architect
- Timeline
- 2026 · Live
$260K/yr
Labor cost absorbed
Energy Sector · Enterprise
An energy company was running three separate teams against three separate workflows: tier-1 customer support, lead qualification, and customer-facing device control. I architected a single multi-agent platform that consolidates all three behind one routing layer. An intelligent classifier reads each incoming request and routes it to the right specialist agent — each with its own tools, CRM integration, knowledge base access, and human escalation path. Outcome: roughly 70% of tier-1 support volume absorbed by AI, around $260K/year in labor cost taken out, faster lead response, and self-service device control for end customers.
- Role
- Lead architect
- Timeline
- 2025 · Live
Curious about a similar consolidation in your business? Book a 15-min call
500+
Shopify merchants served
ChatIn · AI Sales & Support
ChatIn is the multi-channel AI customer support platform I built and ran for Shopify merchants. The system handles inbound across web chat, WhatsApp, Instagram, and Facebook, pulling each store's catalog and knowledge base to answer questions, recommend products, and resolve issues without escalation. At peak it served 500+ stores with a 90% resolution rate — meaning nine of every ten customer conversations completed without a human touching them. The hard engineering wasn't the chatbot. It was the multi-tenant infrastructure, the channel adapters, the smart escalation logic, and the evaluation loop that kept quality stable as the merchant base grew.
- Role
- Lead architect
- Timeline
- 2024 · Live
Live
Government operations
Government Agency · Public Safety
A government avalanche safety agency receives voice reports from field observers all day — noisy, unstructured, time-sensitive. Turning those into formal operational reports was a manual bottleneck exactly when speed mattered most. I built a multi-agent system that converts raw radio transcripts into structured, validated reports in minutes. Four specialist agents run in parallel (weather, snowpack, avalanche, terrain), a deterministic merger combines their outputs, and a critic agent validates everything against domain rules before publication. The reason it actually got deployed — not just demoed — was the three-layer evaluation framework underneath it, which gave operations leadership the evidence they needed to trust AI output in a safety-critical context. Live on Google Vertex AI.
- Role
- Lead architect
- Timeline
- 2026 · Live
Need an eval framework before you deploy? Book a 15-min call
What clients say
Quotes from founders, CEOs, and PMs I've worked with — in their own words.
Amazing work. 10/10.
Dennis was well-prepared and asked great questions during our consultation. Professional, engaged, and easy to communicate with. Would happily work together again.
Dennis is doing an amazing job. He is very knowledgeable and understands what I need.
He knocked it out in a day. Zero bugs, super clean code, clear documentation. I'd gladly recommend him to anyone looking for solid work done professionally.
How we work together
A simple, transparent process — no 90-day strategy decks.
- 01
15-min discovery call
Free. We diagnose, scope, and decide whether I'm the right fit.
- 02
Scoped engagement
Fixed-scope build, retainer, or hourly. Weekly demo cadence. Evaluation harness from day one — no demoware.
- 03
Ship & iterate
Production deploy with you. Hand-off docs. Optional retainer for ongoing reliability work.
About
Principle, denniscode
I'm Dennis. Seven years shipping software, the last three entirely on production AI. I lead AI engineering at a Bay Area consultancy and take on a small number of direct clients in parallel. I work with founders and engineering leadership — no agencies, no offshoring, no 90-day “AI strategy” decks. The work is hands-on, the timelines are tight, and the only thing that matters is whether the system runs in production after I leave.
Stack
Modern stack, vendor-agnostic. I write in whatever your team already writes and deploy on whatever your infrastructure already runs — tools follow the problem, not the resume.
Get in touch
Need AI agents that actually work?
Currently taking on 1–2 new engagements per quarter.