Talk to Your Data: Production-Grade AI Agents for Analytics & Finance
🎬 From Clean Data to Big-Screen Agents: Meet us at Cinema The Pulse! 🎬
Our last meetup tackled scaling data pipelines for pristine data quality. Now that the data is clean, let's make it actionable.
We are thrilled to announce our next meetup on Thursday, June 25th, 2026, hosted by Manychat!
We are taking over Amsterdam's premium Cinema The Pulse! No office rooms this time, architecture diagrams are getting the full, high-fidelity big-screen treatment.
🔍The Focus: Production-Grade Agents
Anyone can prompt a model to generate a basic SQL query in a demo. Building a trustworthy system that safely queries everything from company business metrics to critical treasury data is a completely different challenge. We are peeling back the architecture of real, production systems, including a tool built on Claude Code that lets non-technical users bypass the analyst queue by simply typing /ask.
You will learn the exact engineering patterns needed to transition from prototype to production:
Open Architecture: Designing thin-client plugins, FastAPI services spawning headless Claude, and a "brain" repo for metric definitions.
Smart Routing: Directing plain-English questions to the right business domain and handling multi-turn follow-ups seamlessly.
Security & Isolation: Enforcing multi-tenant data isolation, blocking destructive SQL, and preventing prompt injections.
Reliability Tactics: Taming hallucinated columns with curated views and handling schema drift in highly regulated fintech industries.
Excited to see how production-grade agents actually work under the hood? Space is limited in the theater, so grab your seat now for an evening of high-signal engineering insights, cinematic views, and great networking with the PyData Amsterdam community!
Agenda:
18:00 - 18:55: Welcome with food and drinks! 🍺🍕
18:55 - 19:00: Host company intro - Manychat
19:00 - 19:45: Talk 1: Answers You Can Question: Building a Trustworthy Self-Service Analytics Agent by Alex Litvinov
19:45 - 20:00: Short break
20:00 - 20:45: Talk 2: From chat to insight: reliable AI agents for financial data by Niels Neerhoff
20:45 - 21:30: Networking & drinks
Talk 1 : Answers You Can Question: Building a Trustworthy Self-Service Analytics Agent
By Alex Litvinov
Talk Summary
At Manychat we built an analytics agent on top of Claude Code that lets non-technical employees ask data questions in plain English—no SQL, no analyst queue. A teammate types /ask and a question; the agent clarifies it if needed, routes it to the right business domain, writes guarded SQL against curated views, and returns a formatted answer—multi-turn, so follow-ups just work.
This talk walks the build from prototype to production. We’ll open up the architecture—a thin client plugin, a FastAPI service spawning headless Claude, and a “brain” repo of always-on rules and per-domain metric definitions—and the hard parts: routing questions to the right domain, taming hallucinated columns with curated views, blocking destructive SQL, and scoping data access safely.
Bio
From bioinformatics to backend systems, Alex has spent 15+ years building data-heavy software across healthcare and science. Today he’s a Data Engineer at Manychat, working where platform data engineering meets ML/AI.
Talk 2 : From chat to insight: reliable AI agents for financial data
By Niels Neerhoff
Talk Summary
With the rise of chat interfaces, users increasingly expect to interact with their data in plain English. Large language models make this possible, but deploying a model against financial data introduces risks.
This talk walks through the engineering challenges of building a natural language interface for treasury data: the kind of tool where a CFO can ask, "What's our net cash position this quarter?" and get a trusted answer, without writing any SQL.
We'll cover topics such as multi-tenant data isolation, schema drift, prompt injection, and why you can't rely on the model to enforce access control. Drawing on production experience at Palm, a fintech platform for cash flow forecasting and treasury management, we'll look at the patterns that make NL-to-SQL agents reliable enough for users in a regulated industry. You'll leave with concrete techniques for building agents that are capable, and safer to ship.
Bio
Niels is a software engineer working on data and AI, with an interest in reliable, autonomous systems. He cares about architecture, testing, and automation. When not working, you'll find him on his gravel bike, or listening to all sorts of music.
Directions
📍 Cinema The Pulse, Hildegard Von Bingenstraat 4, 1081 LH Amsterdam, Netherlands Hall #4. Just a 5 min walk from the Amsterdam Zuid train station.