EMBEDDED SF: CraftifAI Physical AI Hackathon
A 10-hour hackathon where you go from a plain-English spec to firmware running on a real dev board, or to a computer vision model doing real-time inference on a GPU, using CraftifAI's AI agents for hardware engineers.
📅 Saturday,Oct 10, 2026 · 9:00 AM – 6:00 PM PT
📍 San Francisco, CA
👤 Individual participation, no teams
🛠 ESP32 boards and sensors provided, cloud GPUs available.
🎒 Bring your own STM family board and sensors, or your NVIDIA Jetson, if you prefer
Why we're running this
AI coding agents changed how software gets shipped, but embedded and edge AI development have barely moved. If you've asked a general-purpose agent for an STM32 peripheral driver, you've probably watched the code compile fine and then fall over on the bench, because the agent guessed the clock tree or the pin mux. Vision pipelines go the same way. Ask twice and you get two different pipelines. One decodes video on the CPU and can't keep up with the camera. The other expects raw frames when your camera sends MJPEG. The agent can see what hardware it's on. It just doesn't know what actually works on it, so it guesses, and the guess changes every time.
CraftifAI builds agents that don't stop at a guess. FirmGen writes firmware for ESP32 and STM32 MCUs, then compiles it, flashes it to the board and debugs it there. PipeGen turns Python CV code and ONNX models into production-ready NVIDIA DeepStream apps, benchmarked on the GPU and packaged in a container. This hackathon is your chance to put both to the test on real hardware, in a single day.
Track 1 · FirmGen: Embedded Firmware
Describe the device you want to build, whether it's a robot subsystem, a wearable or a sensor node. FirmGen plans the firmware architecture, writes the drivers and application logic, then compiles, debugs and flashes it to the board on your desk. ESP32 family dev boards and matching sensors are provided on site. If you prefer to work on STM32, bring your own board and sensors, as we're only providing ESP32 family hardware.
FirmGen also runs on Macs that meet these requirements:
macOS: 26 (Tahoe) or newer
Hardware: Apple Silicon only (M1 or later).
Disk space: About 8 GB free for the app and toolchain.
Bring: a laptop with a USB port. (If you don't have USB A port in your laptop , kindly get a USB hub).
Track 2 · PipeGen: Computer Vision & Edge AI
PipeGen & Edge MLOps: Computer Vision & Edge AI Describe what you want the camera to detect and what should happen when it does, whether it's counting people at a door, spotting a missing hard hat or flagging a car parked where it shouldn't be. PipeGen starts from an existing repo. It can pick the closest match from its catalog of use-case repos, work from one you bring, or use Deep Search to scan through repos and find the best fit. From there it builds the pipeline around your camera, measures how fast it runs and deploys it for real-time inference on your GPU. Need your own model? If nothing out there fits, train one with Edge MLOps. Bring labelled images or use one of its built-in public datasets.
Training runs on a Linux laptop with an NVIDIA GPU or on one of our cloud GPUs, not on a Jetson. Edge MLOps checks that the exported model gives the same results as the trained one, then hands it to PipeGen, which can still deploy it to your Jetson.
Participants working with PipeGen are encouraged to bring their NVIDIA GPU-enabled laptops to the event.
We’ll also provide cloud GPU access for up to 30 participants, available on a first-come, first-served basis. This will give teams additional compute capacity for building, testing, and running their Physical AI projects during the hackathon.
If you need access to a cloud GPU, make sure to claim your spot early.
Who should apply
Firmware and embedded engineers, robotics engineers, computer vision and perception engineers, and hardware founders who want to see how far AI agents can go on real hardware.
Pre-hack office hours
On Tuesday, Oct 12, 7:00 – 9:00 PM PT, accepted hackers join a live session with the CraftifAI engineering team to set up access, test their board or cloud GPU, and get technical questions answered before the day.
Agenda
9:00 – 9:30 AM — Check-in & Hardware Pickup
9:30 – 10:00 AM — Keynote from CraftifAI founders Pratik Sharda (CEO) and Yashwant Dagar (CTO), live from Bengaluru
10:00 AM – 12:00 PM — Hands-on Labs: FirmGen and PipeGen walkthroughs with the CraftifAI team
12:00 – 4:00 PM — Hacking & Testing, with the CraftifAI team online for technical support, working lunch
4:00 – 4:30 PM — Buffer, while we evaluate the projects
4:30 – 6:00 PM — Live Demos & Presentations, Closing Ceremony & Awards
Prizes
FirmGen Track Winner: $250 for 2 winners + 5M CraftifAI Orbit credits
PipeGen Track Winner: $250 for 2 winners + 5M CraftifAI Orbit credits
About CraftifAI
CraftifAI is building the AI engineering layer for physical products. Every robot, drone, smart camera and industrial controller runs on firmware and edge AI software that engineers still write by hand, against datasheets, board revisions and silicon constraints that general-purpose AI doesn't understand. Our mission is to give hardware engineers agents that reason about the hardware the way they do, so a working device is hours away instead of months.
Orbit, our platform, brings together FirmGen for firmware, PipeGen for computer vision and AgentIQ for FPGA design.
EMBEDDED SF is the second edition of our hackathon series. The first was held at Microsoft, Bengaluru.
Spots are limited.