PyData Boston - Cambridge

​
We're back in action with a special three-speaker bonanza for Boston AI Week

We'll start with an intro to building agents in multiple frameworks, move on to considerations around hosting LLMs and wrap up with a dive into responsible AI. Join us!

Do not arrive before 6:30pm!
📅 Schedule:
6:30–7:00 — Networking (doors open)
7:00–7:10 — Introduction
7:10–7:35 — Krithika Murugesan - Building Agentic AI Systems
7:35-8:00 - Sebastian Wallkötter - The physics behind (self-)hosting LLMs
8:00–8:30 - Break
8:30-8:55 — Ananya Sharma - Detecting and Mitigating Bias in Automated Mortgage Lending
9:30 — End

**IMPORTANT** You MUST fill out the registration form accurately in order to get access to the venue. No guests/drop-ins, all unregistered folks WILL be turned away

Krithika Murugesan
Title: Building Agentic AI Systems
Description: In this session, I will introduce the fundamentals of Agentic AI and demonstrate how to build intelligent, autonomous AI systems using modern frameworks such as LangGraph, LangSmith, and CrewAI. Participants will learn how AI agents reason, collaborate, use tools, maintain state, and execute complex workflows through practical examples and architecture patterns.

Sebastian Wallkötter
Title: The physics behind (self-)hosting LLMs
Description: Why do output tokens cost more than input tokens? Why are batch API calls 50% cheaper than normal API calls? Where does the $/M token price come from in the first place? I want to answer this while building a cost model for a local LLM deployment. We'll cover compute, bandwidth, and vRAM dynamics and the effects of K/V cache, context window, quantization, GQA (grouped-query attention), and multi-request batching. Think of this as a (high-level) behind-the-scenes of a per-token LLM API.

Ananya Sharma
Title: Detecting and Mitigating Bias in Automated Mortgage Lending
Description: How can an AI system discriminate without ever seeing protected attributes? This talk explores how automated mortgage models can reproduce historical lending disparities through proxy variables and biased data. It presents practical approaches for detecting and mitigating algorithmic bias using statistical fairness testing, explainability, model governance, and continuous monitoring (edited)