←
Christopher Pietsch664 posts
Christopher Pietsch avatar

Christopher Pietsch

@chrispiecom

interaction designer / creative coder / ML

◷ Joined May 2011▧ 86 Photos and videos
Christopher Pietsch avatar
Christopher Pietsch@chrispiecom·4 Jun 2024

I have been cooking up a project over the last year and now it is finally ready: Merge your creativity with AI and transform the world around you! ✨Transferscope is a handheld device that lets you capture any object or concept and blend it seamlessly onto any scene.

1:06
Brian T. Jacobs avatar
⟳ Christopher Pietsch retweeted
Brian T. Jacobs@btjakes·Jul 24

Big news 🌎🌍🌏: Earthling Maps v1 has launched! Read about it, give it a try, and reach out with any feedback. They're maps that aim to work well even when really small, with defaults to set the stage for great cartography regardless of map-making experience.

Earthling Maps is here! It's all about creating focused maps for articles, newsletters, social media, and presentations, to orient your audience. Learn more in the launch post: earthlingmaps.com/updates/in…

Christopher Pietsch avatar
Christopher Pietsch@chrispiecom·Jun 30

Also moved the UI off Gradio to a small React/Vite workbench: chat on one side, a source panel that fills in as the answer streams on the other. You attach the board photo right in the composer — drag, paste, or pick a file. Plain /api/ask SSE; answers log to JSONL.

Christopher Pietsch avatar
Christopher Pietsch@chrispiecom·Jun 30

The fireworks.ai embeddings API I tried was quietly text-only for a multimodal model. Two different board photos, same text gave cosine 1.0000 — the image was dropped. Self-hosted on the MI300X it was 0.59. That's why I run it myself.

Christopher Pietsch avatar
Christopher Pietsch@chrispiecom·Jun 30

The infra I'm happy with: all three Qwen models on one AMD MI300X — embedder, reranker, and a 35B-A3B MoE. 192GB fits them under vLLM, no quantization. The one gotcha: boot the biggest first, or vLLM won't find a contiguous block for it.

Christopher Pietsch avatar
Christopher Pietsch@chrispiecom·Jun 30

Retrieval is the interesting part. The photo gets embedded too (Qwen3-VL), so it IDs the board from the image before searching. Then vectors, lexical, and a small rule-based entity graph that feeds boards, pins and protocols to the reranker. Every answer is cited.

Brian T. Jacobs avatar
⟳ Christopher Pietsch retweeted
Brian T. Jacobs@btjakes·May 12

I'm long overdue for a social media update, so here it goes. I moved out of DC to the Hudson Valley and I've been freelancing, continuing to enjoy making graphics and tools, designing, coding, and a lot of mapping. I'm about 2 hours north of New York City

Post attachment
Prathyush avatar
⟳ Christopher Pietsch retweeted
Prathyush@prathyvsh·1 Nov 2024

A powerful way to think about user interfaces is as them being bases of a latent space manifold. There is a nascent niche of interesting experiments exploring these and I will try to curate some of these that came into my notice in this thread.

An interface defines the metric of search space

0:27