Back to home

POSTS / CASE STUDIES

Working notes from the product side.

Three evolving accounts of building multi-agent systems, governing AI at scale, and designing growth systems with healthy product economics.

01 · Cleo · 2026–present · Multi-agent AI

working draft

Designing a financial assistant for 5M+ users

The product challenge was not simply to make a chatbot more capable. It was to balance conversational quality, financial-tool accuracy, latency, cost, and the pace at which the team could learn.

The challenge

A single-model approach makes every task pay the same cost and creates a blunt quality ceiling. At financial scale, the assistant also needs dependable tool use and a clear way to evaluate changes before they reach customers.

My leadership

I shaped a multi-agent framework using LangGraph and LangChain, with task allocation and model routing for the right quality-cost trade-off. I also standardised evaluations and strengthened interaction logic and feedback loops so experiments could move faster without lowering the quality bar.

50%MAU engaging with the conversational agent
−30%LLM inference cost
experimentation speed
+20%financial-tool accuracy

02 · Revolut · 2024–2025 · AI governance

working draft

Building safer AI operations at global scale

AI-enabled content and fraud systems create value only when teams can operate them safely, measure them consistently, and improve them without introducing unacceptable risk.

The challenge

The work crossed customer experience, content operations, fraud prevention, and governance. Success depended on connecting technical systems to an operating model that could work across several product lines.

My leadership

I owned product strategy and roadmap alignment, leading two cross-functional teams with 12 people. We introduced an AI governance framework across three product lines, established a North Star metric, and ran more than 50 experiments on content ranking and customer experience.

−50%fraud-related cost
−30%content review time
+15%customer satisfaction
50+A/B tests run

03 · GenAI Co · 2021–2024 · Growth systems

working draft

Scaling a GenAI platform to 3.5M+ monthly users

The goal was durable growth: improve the experience enough to expand usage while strengthening monetisation and unit economics at the same time.

The challenge

Content supply, feed quality, creator behaviour, pricing, activation, and retention were tightly coupled. Optimising one metric in isolation could easily damage another part of the system.

My leadership

As CPO, I led the product system across recommendation ranking, pricing experiments, creator incentives, and activation and retention loops. I built the analytics foundation and used outcome planning and Jobs to Be Done to give teams a shared view of customer value.

3.5M+monthly active users
+120%MAU growth
+25%ARPU, reaching $7
+40%lifetime value
−30%customer acquisition cost