Pooja Bhaumik
Founder @Prehearsal
Pooja Bhaumik, founder of Prehearsal: a speaking coach platform, Google Developer Expert for Flutter, former DevRel at FlutterFlow, and the author of Flutter Essential Training, an exclusive LinkedIn Learning course. She has spoken at over 20 conferences in the past five years, including in-person international events like Fluttercon Berlin, Flutter Conf LATAM, and F3 Prague. She is also passionate about mentoring, especially women in STEM, and has participated in numerous panels on the subject.
AI Observability for LLM-Powered Flutter Apps
Most Flutter apps today are LLM-powered, and it's often just a few steps to bring an agent into the mix. But once your app starts calling models like Gemini or OpenAI, debugging becomes fundamentally different. Unlike traditional APIs, LLMs are non-deterministic, making it difficult to understand why responses change or costs increase.
This talk explores practical ways to add telemetry and observability to AI-powered Flutter apps (that is using cloud backends like Supabase).
Some of the things we'll cover:
- Which AI metrics matter: cost, latency (and why percentiles beat averages), token usage, errors, and output quality
- How telemetry helps you monitor, debug, and improve LLM-powered features
- Hosted AI observability platforms like Langfuse: tracing, evaluations, dashboards, pricing, and alternatives
By the end, you'll have a practical framework for monitoring, debugging, and improving your LLM-powered features.