I take search, data and ML products from legacy to AI.

Upland Software: from keyword search to agentic search

• $5M → $23M ARR
• 65% lower latency
• ~30% lower LLM cost per query
• 12+ AI search releases

Context. A pure keyword enterprise search product inside a Nasdaq-listed knowledge and content management company.

The bet. Go AI search, then agentic search with actionable workflows for Legal, Healthtech, Fintech and US Government.

What I did. Owned roadmap, pricing strategy, SKU updates and vertical positioning. Introduced an AI evaluation framework tracking answer quality, accuracy, latency and LLM cost-to-serve, and made it the release bar. Ran customer and partner advisory sessions.

Outcome. As a team result with engineering and go-to-market, the line grew from $5M to $23M ARR. Across 12+ releases, search latency fell 65% and LLM cost-to-serve per query fell about 30%. My part: the roadmap, the pricing and the eval bar.

What this shows. I can turn an AI feature into a business: a quality and cost bar every release has to clear, and pricing that matches the value.

Canaccede Financial: ML inside a regulated system of record

• ~20% agent productivity
• ~40% faster APIs
• 99.9% uptime
• 15 person team

Context. A Canadian consumer-credit fintech running on a business-critical, tightly coupled financial system of record.

The bet. Migrate in stages to microservices, then put ML where the work happens: a ranking model that tells agents which accounts to work first.

What I did. Led the product side of the staged migration while protecting core financial operations, then embedded predictive models in billing, credit-risk, reporting and resource-allocation workflows. Led a 15-person agile team and aligned engineering, operations, compliance, finance and executives on one roadmap.

Outcome. The team delivered the migration with 99.9% uptime and API responses about 40% faster, and the ranking model lifted agent productivity about 20%.

What this shows. I can change the platform and put ML into production without risking the business that runs on it.

CTO Bees Inc.: from analytics to AI-powered insights

• 12 person product & engineering org
• 3 PMs managed
• 15+ custom solutions

Context. An enterprise analytics and workflow-automation firm, formerly Data Naya.

What I did. Built and led a 12-person product and engineering organization, managing and mentoring 3 Product Managers. Set strategy for analytics and workflow automation (resource forecasting, time capture, billing, capacity planning) and repositioned the offering around AI-powered insights.

Outcome. 15+ custom analytics and automation solutions delivered, with cloud migration and cost optimization planned across AWS, Azure and GCP.

What this shows. I have led PMs as well as products, and I can grow the people behind the roadmap.

PocketBrains: an on-device AI agent for iOS

• 30.2% → 39.6% held-out accuracy
• 24 tools
• 0 cloud calls

Question. For people whose work can't go into a cloud chatbot, is an agent that never leaves the iPhone good enough to run their tasks, projects and notes?

What I built. A 24-tool agent that plans multi-step requests, acts on a local store, answers from notes with citations and can undo anything it did, on Apple's on-device model, an optional open model (MLX), or a deterministic fallback.

Outcome. End-to-end accuracy on a held-out set rose from 30.2% to 39.6%. The test set was frozen before tuning, and every number is published from CI.

What this shows. I still build, and I publish where my own work falls short, not just where it wins.

Most AI features should start as a workflow fix.

Start from the workflow and the metric. If better search or a simple rule moves the number, ship that first and save the model for what only a model can do.

No eval, no release.

Every AI release clears a quality, latency and cost-per-query bar, measured before and after. If we can't measure it, it doesn't ship.

Cost per query is a product decision.

It shapes pricing, packaging and margin long before scale, so it belongs on the roadmap, not just in the infrastructure bill.

Never big-bang a system of record.

Migrate in stages behind the business that runs on it, even when that's slower and less exciting.

A roadmap only product believes in is a wish list.

Engineering, data, compliance, sales and execs agree on one direction, or it isn't the plan yet.