HACKHackers
in the Loop
Join community →

Projects / Active

What’s on the workbench.

Open source software, research tools, GPU and model research, hardware experiments, and the Learning Hub. Hardware listed here is shared across projects; it does not indicate reserved capacity or available allocations.

Updated November 2, 2026.

Explore our GitHub ↗All projects →

Featured / The main work right now

In development

Small Model Arena

Small models, measured the way people really run them. Every quant against its full-precision original.

Read more →GitHub ↗

Open source software

1 project

v0.2.1 released

fxbot

A home for your own bots. One small server and a database run on your machine, and the browser is the app, on desktop and phone. Each bot gets its own look, instructions, model and tools, and talks to a local model server, an API key, or a ChatGPT, Grok or Vercel account. The core stays small: bots, their chats and their files. Skills and MCP servers are add-ons today; plugins that change the app itself, memories, and bots that message each other are next.

Explore fxbot on GitHub ↗Visit fxbot.sh ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Runs on one machine of your own
  • Used every day in the Lab

Research infrastructure

1 project

v0.1.1 nearly ready

Thermite

Open-source autoresearch: AI agents run experiments in a loop, for hours or days, on your own machines. You set the goal: a faster program, a list of checks to pass, or a question to answer. Agents only propose. A separate judge, which they can neither read nor change, builds each attempt, measures it against the best version so far on the same hardware, and keeps only what it confirms. One dashboard, with a REST API and MCP, shows and steers every campaign on every machine.

Explore Thermite on GitHub ↗Visit usethermite.com ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • GPU server lanes
  • Kubernetes work cluster
Working toward
v0.1.0, the first public release, came out on October 26, 2026. v0.1.1 is nearly ready.

GPU & model research

3 projects

v0.1.0 released

tensorless

Custom inference for the Tesla P100, a 16 GB data-centre GPU from 2016 that sells for about $100. One command pulls a model tuned for the card and the exact engine it was tuned with, checks both, and starts an OpenAI-compatible server. The first release runs Qwen3.8-27B in PN3, the Lab's own 3-bit format for the P100, at 30 tokens per second, measured on a real P100. It grew out of the Lab's P100 engine and PN3/PN4 quantization research, and Thermite campaigns put PN3 into the engine and made it fast.

Explore tensorless on GitHub ↗Visit tensorless.dev ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Tesla P100 16 GB
  • Thermite's P100 research lanes
  • DGX Spark for quantization
Working toward
The original goal still stands: at least 31 tokens per second without speculation and 40–60 with it, at 32K and 65K context. PN3's long-context speeds are not measured yet.

In development

Small Model Arena

A benchmark launcher for small models. Give it one Hugging Face checkpoint in the quantization you name; it picks the engine people would really use, plans memory, serves the model and measures it: determinism, perplexity and drift from a full-precision reference, quality suites, GPU kernel writing and deep-research tasks, then speed, while it logs the GPU's clocks, power and temperature. Every run is a resumable record of what someone gets running that model that way.

Explore Small Model Arena on GitHub ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • DGX Spark
  • 4 × V100 in the GPU server
  • Lab machines for grading and code tasks

Active research

Practical Small Model Usage

Getting useful, high-quality results from small models on memory-constrained systems. The research also explores mixtures of models and swarms of task-specific 500M-parameter models.

Project lead
Shannon Duncan ↗
Hardware & resources
  • DGX Spark
  • GPU bench
Working toward
Run 2B–30B-parameter models well on systems with 16 GB of memory or less. These are research goals, not published benchmarks.

Hardware experiments

2 projects

Exploratory research

FPGA model inference

Testing FPGA approaches to encoding model weights and tensor operations directly in hardware.

Project lead
Shannon Duncan ↗
Hardware & resources
  • Microbench / FPGA platforms
Working toward
Explore whether a single card could eventually run a 12B-parameter model at around 13,000 tokens per second. This is a long-term ambition, not an achieved result.

In development

A tiny pet on your wrist

A Tamagotchi-style watch or wristband game Shannon is building with his daughter. She is designing it and working with Claude Code and an ESP32-S3 touchscreen. The separate wrist-agent idea has been folded into this project.

Project lead
Shannon’s Daughter
Hardware & resources
  • Microbench
  • ESP32-S3 with touchscreen

Books & learning

1 project

Live, growing

Learning Hub

Free books and short animated lessons at hackersintheloop.org/learn. It starts with From Models to Custom Silicon, a six-book series that goes from how modern AI models work to building hardware for them. Every chapter can be read on the web, downloaded as a PDF, or watched as short videos with Dabble the duck and Flit the moth.

Visit the Learning Hub ↗
Project lead
Shannon Duncan ↗
Hardware & resources
  • Lab experiments as real examples
  • Writing, voice and animation pipeline
Working toward
Book I, How Modern AI Models Work, has 3 of its 23 chapters published.

Building something of your own?

Share it with the community, ask for feedback, or tell us what you’re learning. You don’t need a project or Lab access to take part.

Connect with the community →

Curious what the Lab is for?

The Lab is where our core projects are built, run and measured. See how it’s built, what it drives and how the work is checked.

Explore the Lab →

Good company for your curiosity

Bring what you’re working on.
Or what you’re wondering about.

You don’t need a project or Lab access to take part.

Join community →