LLM Prototype for Fintech Product Discovery
Blinko AI was an early-stage London fintech startup exploring how AI and gamification could support personal finance products for younger customers. The CTO brought me in to evaluate whether Large Language Models could support a B2B product direction for banks and fintech teams, and to build a functional demo of an LLM-driven assistant inside a banking app.
Problem: The team needed a realistic view of what LLMs could and could not do in a regulated financial context: hallucination risk, privacy constraints, integration with existing product knowledge, and whether the cost profile made sense at startup stage. They also needed something concrete to put in front of potential bank and fintech partners, not a slide deck.
What I built: A multi-agent assistant orchestrated with LangGraph. An intent-classification layer routed each user prompt to the right specialist agent: a text-to-SQL agent that queried the user's transaction data for personal finance questions ("Where have I spent the most this month?"), a financial-education agent for conceptual questions ("What is compound interest?"), and an event-discovery agent that surfaced local activities for the product's youth-oriented audience. The demo ran on a mock SQLite dataset with a sample user, transaction history, and real events scraped from a local listings site, so every answer was grounded and reproducible in front of stakeholders. Tying factual answers to SQL results rather than free generation was a deliberate choice to contain hallucination risk.
Advisory scope: Alongside the build, I compared LLM providers and integration strategies, and translated broad AI ideas into concrete product constraints: what could be demoed, explained, and defended in front of a bank's compliance-minded stakeholders, and what could not yet.
Outcome: Blinko used the demo to land a partnership with an emerging youth-oriented neo-bank. The partner later failed its seed round and shut down before the integration progressed, a common ending in early-stage fintech and outside the scope of this work.