For nearly two decades I've shipped production software for companies that depend on it — Netflix, Mudflap, the New York Public Library. Today I build AI agents and dashboards that can quietly run a mid-sized company's most repetitive operations.
Lucas Cioffi, Principal
I started this practice because I saw a gap that nobody was filling well. Mid-sized companies have real, expensive operational problems — shipping documents that don't match orders, invoices that need three-way reconciliation, internal analytics that wait days for the data team — and AI agents can now solve them. But the expertise to design and ship those systems reliably is still rare.
Big consulting firms move too slow. AI startups don't understand operations. I sit in the middle: a senior production-software engineer who has spent the last two years equipping myself with the latest agent tooling, and who has been shipping Ruby on Rails systems for nearly two decades.
Before this, I was a senior backend engineer at Mudflap, where my code supported millions of dollars in daily fintech transactions. I led teams for Netflix (via DockYard) and the New York Public Library. I co-founded Qiqo, where I built an event platform that grew to 200,000 users and a generative-AI concierge with retrieval augmented generation (RAG) and vector search.
Starting my career, I was an Infantry Captain in the US Army, graduating from West Point and Ranger School. I served as the executive officer for a 125-person headquarters company, and led soldiers on their first combat mission, a three-day, 500-kilometer convoy from Kuwait to Baghdad. I was assigned as base security officer for a forward operating base of 1,200 American soldiers and 100 civilians which experienced elevated rifle, RPG, and mortar attacks.
“That experience is why I build software the way I do — for systems that have to actually work, under real-world stress, with consequences if they don't.
Off-the-shelf tools don't match how your business actually runs. So work falls back on spreadsheets, email threads, and senior people clicking through screens.
Three-way matches. Shipping doc audits. Vendor invoice checks. No one wants to own it. So it stays manual, error-prone, and quietly expensive.
LLM-driven agents are now reliable enough to read documents, cross-reference systems, and flag exceptions — when someone designs the workflow correctly and builds it to last.
I sit with your operations team and trace the actual workflow. Not the org-chart version — the real one, with all the workarounds. We identify the highest-leverage process to automate first.
I map the agent design, data flow, exception cases, and integration points. You get a clear architecture document and an honest assessment of what will and won't work.
A real, running system — Ruby on Rails dashboard, Python agents — handling your actual documents, your actual edge cases. Not a slideware demo.
Production deployment, runbooks, monitoring, and a knowledge transfer to your team. You own the code, the data, and the system from day one.
Designed the architecture for a multi-agent system at a $200M company to automatically verify shipping documents against orders — catching mismatches before they became chargebacks, claims, or customer escalations.
Built a generative AI concierge with retrieval-augmented generation, vector search, and a Rails-based interface, deployed on top of an event platform with 200,000 users. Production-grade RAG on real infrastructure.
Senior backend engineer at Mudflap, building third-party fintech integrations and metrics infrastructure on Rails, PostgreSQL, AWS, and Sidekiq for a system supporting millions of dollars in daily transactions.
An interactive dashboard letting non-technical administrators query a PostgreSQL database in plain English, surfacing trends that previously required a data analyst and a multi-day turnaround.
Orchestrated specialized agents that transform 10,000-line product requirement documents into full-stack web applications — proving the patterns for reliable multi-agent coordination at scale.
Led engineering teams at DockYard (Netflix contract) to build creative partner tooling, and at the New York Public Library to replace the main website search engine. Rails, Ember, React, AWS microservices.
Every engagement is supported by a small team of specialists I trust. They handle design, project management, and delivery rigor — so the systems we ship don't just work, they're actually adopted by the operations teams that need them.
Sandra leads design and project delivery — translating messy operational reality into interfaces that real teams adopt on day one. She partners with me on every client engagement to ensure scope is clear, work ships on schedule, and the dashboards we build are ones operations teams actually want to use.
The technology choices that survive your IT review, your security review, and five years of operations.
I take on a small number of engagements at any one time. If you have a repetitive operational process that's costing real money and real attention, send me a note describing it. A first conversation is always free and always direct.