@apreshill

Leading product @PixeltableHQ

apreshill.com
Joined April 2015
  • Tweets7.8K
  • Following2K
  • Followers14.4K
1.4KMedia
Alison Presmanes Hill retweeted
Pixeltable
@pixeltablehqOct 16
If you've been searching or asking your agent "how do I <insert multimodal AI> in #Python?", check out our new how-to guides covering: → How do I extract keyframes from video? → How do I build video RAG? → How do I build a video processing pipeline? → How do I build a multimodal RAG pipeline? → How do I chunk PDFs for RAG? → How do I embed images for search? → How do I do semantic search? → How do I do hybrid search? → How do I build an AI data pipeline? → How do I add computed columns with LLM calls? pixeltable.com/learn/how-to
1
3
167
Alison Presmanes Hill retweeted
Pixeltable
@pixeltablehqOct 15
We're building an app a day with Pixeltable. App Nº 4 helps people keep a daily habit, labeling each streak. At every check-in, Pixeltable recomputes the streak's label with plain Python. If an AI model wrote each streak a note, it'd be recomputed too. github.com/pierrebrunelle/pi…
Alt: This app helps people track their habits and streaks. Stage 1, a habit is saved: habit Meditate, category health, streak 7 days. Stage 2, Pixeltable computes: streak label on a roll. Stage 3, the API serves the results: POST /habits saves the habit; POST /habits/update, a check-in at 30 days returns "unstoppable"; GET /habits/by-category, one category's habits, longest streak first.

ALT Alt: This app helps people track their habits and streaks. Stage 1, a habit is saved: habit Meditate, category health, streak 7 days. Stage 2, Pixeltable computes: streak label on a roll. Stage 3, the API serves the results: POST /habits saves the habit; POST /habits/update, a check-in at 30 days returns "unstoppable"; GET /habits/by-category, one category's habits, longest streak first.

1
2
134
Alison Presmanes Hill retweeted
Pixeltable
@pixeltablehqOct 13
We're building an app a day with Pixeltable. App Nº 3 is a recipe backend that finds meals you can cook in the time you have. Pixeltable recomputes the ingredient count on every edit. If you added an AI-generated recipe summary, for example, Pixeltable would recompute it too. github.com/pierrebrunelle/pi…
Alt: This app keeps home cooks' recipes and finds quick meals. Stage 1, a recipe is saved: title Lemon pasta; ingredients spaghetti, lemon, parmesan, butter; cooking time 15 min. Stage 2, Pixeltable computes: number of ingredients 4. Stage 3, the API serves the results: POST /recipes saves the recipe; POST /recipes/update, add black pepper, and the count goes from 4 to 5; GET /recipes/quick, recipes of 20 minutes or less, fastest first.

ALT Alt: This app keeps home cooks' recipes and finds quick meals. Stage 1, a recipe is saved: title Lemon pasta; ingredients spaghetti, lemon, parmesan, butter; cooking time 15 min. Stage 2, Pixeltable computes: number of ingredients 4. Stage 3, the API serves the results: POST /recipes saves the recipe; POST /recipes/update, add black pepper, and the count goes from 4 to 5; GET /recipes/quick, recipes of 20 minutes or less, fastest first.

1
1
1
177
Alison Presmanes Hill retweeted
Pixeltable
@pixeltablehqOct 13
We're building an app a day with Pixeltable (but the building bots work weekends so we are BEHIND 😃 ) App Nº 2 is a bookmarks backend. Someone adds a third tag to a link, and Pixeltable recomputes the tag count, now 3, with a plain Python function. An AI model call to generate a topic, for example, would be recomputed too. github.com/pierrebrunelle/pi…
Alt: This app saves your links and groups them by website. Stage 1, you save a link: link https://fastapi.tiangolo.com/tutorial/, tags python,web, title "FastAPI tutorial: first steps and beyond". Stage 2, Pixeltable computes: website fastapi.tiangolo.com; number of tags 2; title, cut short, "FastAPI tutorial: first steps a…". Stage 3, the API serves the results: POST /bookmarks saves the link; POST /bookmarks/update, add a third tag, and the count goes from 2 to 3; GET /shelf, saved links for the website it's given.

ALT Alt: This app saves your links and groups them by website. Stage 1, you save a link: link https://fastapi.tiangolo.com/tutorial/, tags python,web, title "FastAPI tutorial: first steps and beyond". Stage 2, Pixeltable computes: website fastapi.tiangolo.com; number of tags 2; title, cut short, "FastAPI tutorial: first steps a…". Stage 3, the API serves the results: POST /bookmarks saves the link; POST /bookmarks/update, add a third tag, and the count goes from 2 to 3; GET /shelf, saved links for the website it's given.

1
1
1
140
Alison Presmanes Hill retweeted
Pixeltable
@pixeltablehqOct 13
We're building an app a day with Pixeltable. App Nº 1 is a sticky-notes backend. Pixeltable fills in every note's first line, whoever writes it: the API or a script. A plain Python function picks it. An AI translation you added would be filled in too. github.com/pierrebrunelle/pi…
Alt: This app keeps sticky notes and finds them by tag. When you save a note, the app picks out its first line. Stage 1, a note is saved: title Groceries; tag home; note body, two lines, oat milk, lemons, basil, then maybe sourdough. Stage 2, Pixeltable computes: first line, oat milk, lemons, basil. Stage 3, the API serves the results: POST /notes saves the note; GET /notes/by-tag returns one tag's notes, newest first.

ALT Alt: This app keeps sticky notes and finds them by tag. When you save a note, the app picks out its first line. Stage 1, a note is saved: title Groceries; tag home; note body, two lines, oat milk, lemons, basil, then maybe sourdough. Stage 2, Pixeltable computes: first line, oat milk, lemons, basil. Stage 3, the API serves the results: POST /notes saves the note; GET /notes/by-tag returns one tag's notes, newest first.

1
1
1
156
We've been busy @pixeltablehq 😅 Check out our latest release to get up to speed on the biggest change we've shipped to our CLI: a whole new workflow that stops agents from "freelancing" your app's backend (and bungling it).
Pixeltable
@pixeltablehqOct 9
New in Pixeltable for August–September: the backend your coding agent builds with. Full recap: pixeltable.com/blog/pixeltab…
The backend your coding agent builds with. Code for app.py and the terminal in 3 numbered parts: 1, The backend as Python classes; 2, A plan before every change; 3, Routes from the same CLI.

ALT The backend your coding agent builds with. Code for app.py and the terminal in 3 numbered parts: 1, The backend as Python classes; 2, A plan before every change; 3, Routes from the same CLI.

1
4
451
Whoops - got too silly building apps for @pixeltablehq from Bozeman 🤠🤠🤠🤠
Justine Moore
@venturetwinsOct 5
I need to know wtf is going on in Montana
2
466
Open invitation: I'll help you think through this multimodal RAG interview, if you let me pitch you @pixeltablehq so you don't have to (re)build all this plumbing from scratch when you get the job 🍞 pixeltable.com/toast

Toast my token budget. Keep the code.

Build and deploy a multimodal app with Alison from Pixeltable. Images, video, audio, or documents. One hour, her AI tokens, and a GitHub repo with the code to keep.

pixeltable.com
Viv
@Vtrivedy10Oct 1
interview question I like - multi-modal RAG w/ an agent can go high level & also rabbit hole into context engineering, tokenization, harness design, search, how models are trained, etc Setup: there’s a folder with multi-modal data - markdown files, PDFs, image files, video files Let’s build a Q&A system over this data. Rabbit Holes & Follow-Ups: - How would a ReadFile() tool work per modality? Let’s talk about preprocessing, tokenization & ingestion into the context window and if they differ per modality? Anything to be careful of? - How would you design the search harness? say we care a lot of about reducing costs from unnecessary reads? or we care a lot about recall - What if we had A LOT of video data, how would your approach change? Would you consider preprocessing (ex: transcripts)? how would that match to how the model builder recommendation based on how it was trained? - what model(s) would you use and why? if multiple, how would they work together in harness? - here are some query patterns. the data is mostly about sports events (transcripts, gameplay footage, stills, reports). most people ask questions about when an event happened. some ask “how” it happened in the real world. let’s talk about how to research and design around these patterns by looking through our data you can go infinitely deep here. i’d like talking through mental models around model context & the harness for preparing context. also details around how to do efficient search and when to pull context into the context window vs operate over it via code very useful to talk through different levels of abstraction with someone, can tell pretty quickly how much they’ve considered agents and the data they operate over, also would just be a pretty fun convo!
1
1
3
1,014
In case you missed it, this is a great prompt for writing a good ticket whether or not you use @linear. I do this during every video tutorial I create for the development team @pixeltablehq.
Karri Saarinen
@karrisaarinenSep 24
Replying to @justpw @linear
Not sure how it exactly happens in your case. 1. we use linear agent a lot to create issues from slack/elsewhere so it's already tuned to do in less verbose way 2. There is the option to run loops to new issues to somehow refactor them to me more human readable 3. Internally some people have this pattern/skill which tells the agent to write the main thing at the top, and then "agent notes" at a collapsable section below. "Write the issue so the main description is useful on its own and human readable. Lead with the goal, the user-facing or product impact, and the key requirements or acceptance criteria. Keep it concise and avoid implementation detail there.Put technical investigation, API references, edge cases, rollout considerations, and source links in a collapsible section titled Agent notes. Organize those notes with short headings when helpful. Don’t repeat the main description inside the section.Preserve important caveats and dependencies, but make the issue easy to scan before expanding the notes."
487
This is the best reason to use @pixeltablehq - our customers start with their data model and a Pixeltable schema that includes multimodal types. Then you can start hurling your tokens into your tables.
Thariq
@trq212Jul 17
building prototypes of mockups, schemas, data models, proof of concepts, etc. is the best way to avoid spending tons of tokens before realizing you don't want the output
2
1,286
Very excited for this new launch from @reve - their models are first class and produce seriously high-quality generations. If you want to play with their models in #Python, here is how to do it in @pixeltablehq docs.pixeltable.com/howto/pr… youtube.com/watch?v=aNEyy17v…
Reve
@reveJun 3
Today, we’re launching Reve 2.0, the best 4K image model in the world. We invented a new way to generate and edit any image using precise layouts. For the first time, it’s possible to create images you can touch.
2
16
3,465
I broke Google Gemini yesterday and got a glimpse into its internal monologue. Strong Bart Simpson energy. Claude counted ~4,000–5,000 repetitions of "I will output the response." We laughed. Good times. 🤣 #AIFails #AdventuresinAI
1
5
669
more replies
1
208
163
If you are into multimodal AI (or just multimodal curious), March was a good month 🍀 You can now generate images (@bfl_ml FLUX), edit them (@OpenAI inpainting/outpainting), resize videos, concat clips, and embed it all (@GeminiApp multimodal) — in the same @pixeltablehq pipeline. 🧵
1
2
4
471
more replies
🎥 Primetime Workshop — 3-part YouTube series from our January workshop with @hugobowne and Vanishing Gradients Acts 1 & 2 live, Act 3 coming soon Watch: youtube.com/playlist?list=PL… Code: github.com/pixeltable/pxt-pr…

Building Multimodal AI Workflows with Pixeltable

Production masterclass series co-hosted with Vanishing Gradients. Learn how to engineer advanced multimodal AI data pipelines for video and audio processing ...

youtube.com
2
2
214
Also shipping: 🔲 Bounding box UDFs — crop, draw, scale, transform 📚 Migration guides for LangChain, LlamaIndex users Changelog: docs.pixeltable.com/changelo… Compare: github.com/pixeltable/pixelt… #MultimodalAI #Python #OpenSource #GenerativeAI

Changelog - Pixeltable Documentation

Release history and updates for Pixeltable

docs.pixeltable.com
137
Load more