Production AI that shipped and stayed up
Thirteen engagements, each on its own page: the pain it started from, how the solution works step by step, and why it holds up. Client names stay confidential.
Expert-Vetted top 1% · 100% Job Success Score (source: Upwork profile, October 2026) · Hong Kong & London
Big 4 client · fully on-premise
Enterprise RAG Chatbot for a Big 4 Accounting Firm
Multimodal retrieval over a Big 4 firm’s full compliance library, tuned for citation instead of summary — every answer arrives with its source paragraph attached.
- Juniors reading thousands of pages per question
- Seniors acting as search engines
200k calls/month · 67% booking rate vs 77% human
24/7 AI Agent Call Center: 200k Calls a Month Across 60 US HVAC Brands
Voice agents orchestrating 200,000 calls a month across 60 US HVAC brands for a 5,000-person enterprise, booking 67% of calls against 77% for human agents — within 10 points of a trained team, around the clock. Built as a product, an AI front desk that answers, books and routes, not a pilot.
- Night and weekend calls unanswered
- Staffing every shift hard to justify
US$1M collected in the first 3 months
AI Payment Collection and Reminders
The companion product to the AI front desk: it chases unpaid invoices with polite, timed reminders and a payment link, and stops the moment the customer pays or replies. It collected over US$1 million for the client in its first three months.
- Chasing payment is awkward and late
- Billable time lost to collections
Ask the EDI in plain English
Natural-Language EDI for a US EDI Solutions Vendor
Business users ask questions of raw X12 — “what are the net terms on invoice 4471?” — and get answers in plain English, from a Llama 3 model fine-tuned on EDI and grounded in the parsed transactions.
- Business and EDI teams speak different languages
- Rejected transactions decoded by hand
Same question, same SQL, same answer
Deterministic Natural-Language Search for a Private-Markets Data Platform
Deal teams ask “industrials in the southwest received this year with LTV under 70%” and get an exact, explainable table — the model plans the query, but never writes free-form SQL.
- 100+ columns only an analyst can query
- Vernacular terms are not column names
Ask in Slack, answered from Confluence
Slack Knowledge Assistant over Confluence for an EdTech Company
An assistant inside Slack that answers staff questions from the company’s Confluence — routed by an agent that decides, per question, whether to search the wiki, the web, or just answer.
- Knowledge buried in Confluence
- The same questions answered every week
Prototype · three rewrite strategies benchmarked
Query Rewriting for a Global Asset Manager’s Research RAG (Prototype)
An asset manager’s internal research assistant retrieved well when users asked well. We benchmarked three ways to rewrite the questions they actually ask, and matched each to the kind of question it fixes.
- Vague questions without keywords
- Three questions asked at once
Nothing leaves the room
Super Chain: Private On-Prem AI for a Family Office
An LLM with retrieval over financial research, macro data and portfolio holdings, with the model chosen per office, running on the office’s own infrastructure. Every answer is cited.
- Documents that cannot leave the office
- Every AI product needed a third-party API
247 hours over three months · 5.0 rating
Voice AI Product Build
A commercial voice product run as a long hourly engagement rather than a fixed-scope prototype — 247 hours across three months, rated 5.0.
- Demo calls are not real calls
- Failures only appear at volume
246 hours · sustained engagement
Multimodal RAG Application
Retrieval over source material that is part text, part table and part image — where naive chunking quietly returns answers that are confident and wrong.
- Tables split mid-row
- Numbers hidden in images
Rated 5.0 · shipped to a live product
Natural-Language Query Layer for Property Search
Users describe the property in a sentence; the system resolves it into the structured query underneath — against a schema that has to return exact results.
- Users translate wants into filters
- Search must return exact matches
Nearly 130 GitHub stars · MIT licensed
N8N2MCP — Open Source
Turns an n8n workflow into a Model Context Protocol server, so automation a team already built becomes a tool Claude or Cursor can call directly.
- Teams rebuild tools they already have
- Integrations duplicated for MCP
Rated 5.0 · advisory engagement
Fractional CTO: ML Architecture & Roadmap Review
Reviewing the machine-learning plan, the stack and the sequencing before the team committed budget to it. Most of the value is in what comes off the roadmap.
- A roadmap never reviewed by someone who shipped ML
- Planning mistakes surface two quarters later
Your problem is probably on this page.
Retrieval that has to cite its sources, an agent that has to stay up, or AI that cannot send your documents anywhere. Tell us which one and we'll tell you what it takes.
Book a Call