# builds Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/builds
> Markdown URL: https://aitinkerers.org/technologies/builds.md
> Technology record last updated: 2026-02-25T05:32:24Z
> Generated: 2026-09-22T04:47:36Z

A premier community platform where PC enthusiasts showcase custom hardware, document build logs, and compete for global rankings.

Builds.gg connects over 100,000 builders who share high-resolution galleries of their custom rigs (from compact ITX setups to $15,000 liquid-cooled towers). The site functions as a technical archive: users list specific components from brands like NVIDIA and Noctua while tracking project milestones. It hosts regular 'Build of the Month' contests and provides a searchable database of thousands of unique configurations. This platform turns individual hardware projects into a collaborative, visual encyclopedia of modern computing performance.

- Official technology site: https://builds.gg
- Public AI Tinkerers demos and talks: 1
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Information Crumbs to Nuggets: Incorporating Humans into LLM-Judges](https://manchester-nh.aitinkerers.org/talks/rsvp_qB5PHlwmBIQ)

We built a prototype of a human-AI workflow for curating the evaluation nuggets that nugget-based LLM judges depend on. The core problem is that incorporating humans into LLM-judge evaluation is usually all-or-nothing, either fully manual or fully automatic, when the reliable answer sits somewhere in between. Single-prompt LLM judges are vulnerable to circularity, where the judge rubber-stamps a system that shares its own blind spots. Nugget judges defend against this, but only if the evaluation artifacts are genuinely human, not LLM proposals that a human merely vets. The catch: authoring nuggets by hand is tedious, and showing the annotator an LLM's proposed nuggets primes and anchors them, which quietly erodes the accountability you were trying to protect. So the prototype inverts the usual flow. Instead of a human phrasing nuggets directly (today's standard) or signing off on LLM-proposed ones, the human acts first and leaves "information crumbs" on the raw data: span highlights of what matters, free-form notes on what is missing, and good-versus-bad judgments. The LLM then converts those crumbs into a canonically phrased nugget. A feedback loop applies that nugget back over a sample of the data and shows the human the impact of the formulation, so they can step in and correct it. The human owns the substance and the judgment, where the work is intellectually demanding; the LLM handles the phrasing and the boring matching, so the annotator is enabled rather than buried in sludge. Live, I will walk through the loop in the prototype: a human highlighting and annotating raw answers, the LLM turning those annotations into nuggets, the nuggets being applied back over a sample, and the human correcting based on the observed impact. I will show the interface, the workflow, and the a trace of what the human contributed from what the model generated.

- Event context: AI Tinkerers Manchester (Bedford), NH - June 2026 Meetup — 2026-06-17 — Manchester NH
- Public talk page: https://manchester-nh.aitinkerers.org/talks/rsvp_qB5PHlwmBIQ

## Related Technologies

- [auto-judge-annotate](https://aitinkerers.org/technologies/auto-judge-annotate) ([Markdown](https://aitinkerers.org/technologies/auto-judge-annotate.md)) — 1 public demo
- [gpt-oss-120b](https://aitinkerers.org/technologies/gpt-oss-120b) ([Markdown](https://aitinkerers.org/technologies/gpt-oss-120b.md)) — 2 public demos
- [LLM-Judge](https://aitinkerers.org/technologies/llm-judge) ([Markdown](https://aitinkerers.org/technologies/llm-judge.md)) — 1 public demo
- [Open AI](https://aitinkerers.org/technologies/open-ai) ([Markdown](https://aitinkerers.org/technologies/open-ai.md)) — 1 public demo
- [OpenAI API compatible LLM endpoints](https://aitinkerers.org/technologies/openai-api-compatible-llm-endpoints) ([Markdown](https://aitinkerers.org/technologies/openai-api-compatible-llm-endpoints.md)) — 1 public demo
- [Software](https://aitinkerers.org/technologies/software) ([Markdown](https://aitinkerers.org/technologies/software.md)) — 2 public demos
- [TREC Auto-Judge](https://aitinkerers.org/technologies/trec-auto-judge) ([Markdown](https://aitinkerers.org/technologies/trec-auto-judge.md)) — 1 public demo
- [URL](https://aitinkerers.org/technologies/url) ([Markdown](https://aitinkerers.org/technologies/url.md)) — 2 public demos