Open Source Legends on data-driven A.I. agents and LLM-assisted PCBs

Two open-source legends will be presenting at our February O‘ahu A.I. meetup:

• From PCB to Production: Developing Embedded Systems with LLMs
• A Data-Driven Approach to Balancing Quality, Latency, and Cost in AI Agents

If you care about:

• Bridging AI software and real-world hardware
• Building agents that don’t silently regress
• Understanding what your AI systems are actually doing
• Closing feedback loops between production and evaluation

This meetup is for you.

🎙️ Talk 1
From PCB to Production: Developing Embedded Systems with LLMs
by Jonas Schnelli

Estimated time: 20 minutes
LLMs aren’t just changing software — they’re reshaping hardware and embedded development.
Jonas will walk through how modern AI tools are influencing the entire embedded workflow, from circuit boards to firmware to production devices.

What is a PCB?
PCB stands for Printed Circuit Board. It’s the physical board that holds and connects electronic components (chips, resistors, capacitors, microcontrollers) using etched copper traces — the backbone of almost all modern electronics, from laptops and phones to ESP32 and IoT devices.

***

🎙️ Talk 2
A Data-Driven Approach to Balancing Quality, Latency, and Cost in AI Agents
by Nathan Marz

Estimated time: 60 minutes
LLMs are inherently unpredictable. A prompt change that fixes one case often breaks another you didn’t test. When the input space is effectively all of human language, ad-hoc testing leads to fragile agents that fail under real-world usage.
Nathan will show how to build reliable AI agents through systematic, data-driven, iterative development. You’ll see how to optimize across the quality-latency-cost tradeoff using production data, controlled experiments, and deep observability.
This talk covers:

• Structured datasets with real inputs and “golden truth” outputs
• Avoiding circular logic when using LLMs to evaluate LLM output
• Why evaluating individual steps of agent workflows matters
• Tracing and telemetry requirements for AI agents
• Automatically collecting datasets from production traffic
• Online evaluation, alerting, and continuous feedback loops

***

About Jonas Schnelli
Jonas has been writing code for 30 years, spanning web apps, native iOS and Android, and Mac applications. He spent 8 years deep in Bitcoin Core’s C++ internals before focusing on hardware engineering and embedded firmware.
Now based in Hawai‘i, Jonas designs ESP32 devices and explores how AI tools are transforming embedded systems development.
GitHub: https://github.com/jonasschnelli

***

About Nathan Marz
Nathan lives on O‘ahu and runs Red Planet Labs, a distributed company building Rama, a developer platform for scalable backends with far less code and infrastructure.
He created several influential open-source projects, most notably Apache Storm, one of the first systems to enable reliable large-scale real-time data processing. Nathan previously worked at Twitter, where he helped start core infrastructure teams, and he is the author of Big Data: Principles and Best Practices of Scalable Realtime Data Systems.
Outside of work, Nathan enjoys hiking, stand-up comedy, and old movies.
GitHub: https://github.com/nathanmarz

***

🌺 This event is sponsored by Hawaii Coworking (https://hawaiicoworking.co) , Darshaun Nadeau (https://flowingblue.com) and Kevin Riggen (https://oahu.ai) .