# Llama-4 Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/llama-4
> Markdown URL: https://aitinkerers.org/technologies/llama-4.md
> Technology record last updated: 2026-02-24T08:32:08Z
> Generated: 2026-09-22T03:43:43Z

Llama 4 is Meta's open-weight, multimodal AI model series: it leverages a Mixture-of-Experts (MoE) architecture for superior efficiency and performance.

Llama 4, released by Meta in April 2025, represents a significant architecture shift to Mixture-of-Experts (MoE), boosting efficiency and performance. The initial release includes two primary models: Scout and Maverick. Scout features 17 billion active parameters (109B total) and a 10 million token context window, optimized for single-GPU deployment. Maverick scales up with 17 billion active parameters (400B total) and 128 experts, delivering competitive results against market leaders. Both models are natively multimodal (text and image input) and multilingual, making them a powerful, open-weight foundation for a wide range of advanced AI applications.

- Official technology site: https://llama.com
- Public AI Tinkerers demos and talks: 2
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Structured Outputs & Batch Processing w/AI](https://nyc.aitinkerers.org/talks/rsvp_G38IYdS2Bb8)

A short overview of some code that does batch processing and when and where you'd want to build it into your workflows. Structured outputs bring their own development challenges, I'll speak to a couple of things I've noticed from using structured outputs over the last 12 months.

- Event context: Building AI Agents with Google Cloud AI — 2025-06-25 — New York City
- Public talk page: https://nyc.aitinkerers.org/talks/rsvp_G38IYdS2Bb8

### [Using ML for Government Transparency](https://nyc.aitinkerers.org/talks/rsvp_c-_XTku9cAI)

Journalists, academics, and advocates can request government documents under the Freedom of Information Act (FOIA) — but these requests often take months or years to fulfill. While talking with 100+ government employees who process those requests, we learned (1) the biggest blocker to a government agency fulfilling a journalist's FOIA request is request quality, and (2) the vast majority of requests these employees process are terrible. So we built a tool to help journalists draft and submit high-quality, easy-to-process requests; it's currently in beta being used by two dozen journalists (incl. from WIRED, POLITICO, NPR) who have filed ~300 requests. I'll be presenting the tool! We trained a small, BERT-based classifier to rate draft FOIA requests along three dimensions: clarity, scope, and (legal) compliance. When the user rates a draft request, we call a large language model (Gemini 2.5 Flash) that suggests improvements to increase the requests's score on a dimension, which the user can accept with a click. The user then submits the perfected request through our platform alongside an LLM-tailored fee waiver (which has been granted 100% of the time so far). (We also have a "brainstorm" flow chaining calls to Llama 4 via Groq & Gemini 2.5 Flash/Pro where a user can input an article, press release, or other FOIA fodder and receive three high-quality request ideas.)

- Event context: Demos and Dim Sum with Deel and Apollo — 2025-06-03 — New York City
- Public talk page: https://nyc.aitinkerers.org/talks/rsvp_c-_XTku9cAI

## Related Technologies

- [AWS Nova](https://aitinkerers.org/technologies/aws-nova) ([Markdown](https://aitinkerers.org/technologies/aws-nova.md)) — 1 public demo
- [Claude-3](https://aitinkerers.org/technologies/claude-3) ([Markdown](https://aitinkerers.org/technologies/claude-3.md)) — 110 public demos
- [Flask](https://aitinkerers.org/technologies/flask) ([Markdown](https://aitinkerers.org/technologies/flask.md)) — 24 public demos
- [Gemini](https://aitinkerers.org/technologies/gemini) ([Markdown](https://aitinkerers.org/technologies/gemini.md)) — 188 public demos
- [LangChain](https://aitinkerers.org/technologies/langchain) ([Markdown](https://aitinkerers.org/technologies/langchain.md)) — 445 public demos
- [OpenAI](https://aitinkerers.org/technologies/openai) ([Markdown](https://aitinkerers.org/technologies/openai.md)) — 112 public demos
- [OpenCV](https://aitinkerers.org/technologies/opencv) ([Markdown](https://aitinkerers.org/technologies/opencv.md)) — 26 public demos
- [Sonnet](https://aitinkerers.org/technologies/sonnet) ([Markdown](https://aitinkerers.org/technologies/sonnet.md)) — 25 public demos
- [Transformers](https://aitinkerers.org/technologies/transformers) ([Markdown](https://aitinkerers.org/technologies/transformers.md)) — 148 public demos