Fonzi’s July Tech Talk

Join us for the July edition of the New York AI Engineers Tech Talk, a monthly gathering focused exclusively on AI engineering. Whether you’re a seasoned machine learning engineer or a full stack dev curious about how to work with LLMs, this is your space to dive deep into real-world projects, technical challenges, and creative solutions.

Each month, we invite AI engineers to share something they’ve built, the challenges they faced, and how they solved them. It’s all substance, no fluff, designed for engineers by engineers.

This Month’s Presenters:

Vinoo Ganesh, CEO & Co-Founder @ Kepler The Model Never Held the Number: Harness Programming for Reliable Agents LLM agents are good at deciding and bad at carrying: ask a model to pass an exact value between tool calls and it eventually drops a digit, which in financial research is fatal. At Kepler we fixed this by splitting the system in two, a non-deterministic model that only decides and a deterministic harness that holds, moves, and computes everything else, so the model works with references to data it never sees. Using a worked example of building a financial statement from SEC filings, John will show how this turns whole classes of agent error from unlikely into impossible, a pattern that applies to tool-using agents in any domain.

PromptQL Building a Company Brain That Organizes Itself: A Shared Context Layer for AI Agents Retrieval is only part of the challenge, shared context is what enables AI agents to perform reliably. This talk explores a self-organizing, Wikipedia-style knowledge layer that helps agents understand evolving organizational knowledge beyond traditional vector databases. Learn how intelligent page selection, link-based retrieval, and human-approved AI learning keep context accurate and up to date, while reducing ambiguity and stale knowledge. Attendees will leave with a practical approach to building AI agents with durable domain knowledge without fine-tuning or massive context windows.

Edis Gonuler, Core Engineer @ Viridium Partners Modernizing Private Market Liquidity with AI Private market transactions often involve complex processes and limited transparency. Viridium Partners will share how they're using AI and automation to help limited partners navigate private market liquidity, streamline buyer discovery, and improve transaction execution. The talk will explore modern approaches to private market advisory, AI-powered workflows, and how technology is transforming investment processes.

The Crowd: You’ll be joining a mix of software developers, data scientists, AI engineers, and technical founders working at or curious about the future of applying AI at the application layer. It’s a casual, high-signal environment perfect for learning something new and meeting peers in the NYC AI Engineering scene.

Agenda: 6:00 to 6:45 PM – Networking with pizza, beer, & wine courtesy of Fonzi AI 6:45 to 6:50 PM – Opening remarks 6:50 to 7:35 PM – Three 10 minutes Tech Talk