Top Tweets for #MetaAIHackathon
π Iβm building Tataloka for the Meta AI Developer Hackathon!
Tataloka uses AI to help people understand and analyze their spaces.
Iβd really appreciate your support! π
π³οΈ Vote: https://t.co/Nu7Myn155k
π§ͺ Try it: https://t.co/s849OecOU9
#MetaAIHackathon #BuildWithAI
Final day of the window: show, don't code. Film a protocol once by hand, get a robot protocol that passes Opentrons' own software. Thanks for following the build. @AIatMeta @cerebral_valley
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
@JulianGoldieSEO Built for the Meta Global AI Developer Hackathon β one memory layer for every AI, local-first and scrubbed before storage. #MetaAIHackathon #CerebralValley #MuseSpark #BuildForYourWorld
Last call before judging: if a robot that learns protocols from a video of your hands sounds useful, a repost helps this solo build a lot. Demo, code and live app are all in the post below.
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Thank you @AIatMeta and @cerebral_valley for the models and the hackathon, @baseten @modal @FireworksAI_HQ for the infrastructure, and the BioVL2 team at Kyoto University for open lab videos.
#MetaAIHackathon
https://t.co/AzpoztARVn
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Impressive....
Glad to see young people taking tech seriously
Build build and buildβ
https://t.co/v42mx1CSZu This link is our Eco study collaboration with Meta AI discusses Eco concerns in the Mid-Ohio Valley #MetaAI #MetaAIHackathon
If you run an Opentrons Flex, I'd love a real-world test: film one protocol you already do, upload it, and tell me what it got wrong. That's the next step this project needs. @opentrons
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Hardest bug: a receipt read while its Bitcoin proof was being written saw a half-written file. Fix: write every proof atomically. Reproducibility tooling has to be boringly correct.
#MetaAIHackathon
https://t.co/AzpoztARVn
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Favorite moment of the build: the robot-server reporting 'succeeded' and every one of 240 commands matching its own analysis, on a protocol that started as a GoPro video of someone's hands. @opentrons
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Try it without signing up: https://t.co/wPp3jQ8uXk has eight samples, seven filmed PCR and miniprep videos and one https://t.co/B2mT1n2J9Q protocol. Click a step and watch the robot do the same thing.
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
CBC will be the end of many kids. Some kids are in grade 10 and they donβt know how to read.
CBC might be good but how it was implemented was wrong. Our resources canβt sustain this method of learning. Change or adopt the @LoWisadev
#MetaAIHackathon
Everything is open source, from the deterministic codegen to the audits: https://t.co/rBjcT5JK7r. Built solo in 9 days for the @AIatMeta hackathon by @cerebral_valley.
#MetaAIHackathon
https://t.co/AzpoztARVn
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
The 2:53 demo video is recorded live from production with Playwright, narrated by an AI voice with word-timed subtitles and assembled in Remotion. Every number on screen is a real result: https://t.co/dPkzIErb7c
#MetaAIHackathon
https://t.co/AzpoztARVn
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Who it's for: labs that bought a liquid handler and use it for one protocol because nobody has time to write Python. Film what you already do by hand; review it next to the footage; send it to the robot.
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
@bowora_com FinePrint β I measured 14 coding agents against their own rules. Built it for the #MetaAIHackathon, code is rough and I'm refactoring next week. 13 of 14 read a file they were told not to touch. https://t.co/VymtczR0wK
@anupamrjp FinePrint β I measured 14 coding agents against their own rules. Built it for the #MetaAIHackathon, code is rough and I'm refactoring next week. 13 of 14 read a file they were told not to touch. https://t.co/VymtczR0wK
Why open weights matter here: the reviewer agent runs on Muse Spark Open, so a lab could run it on-premise next to its robot, with its videos never leaving the building. @AIatMeta @baseten
#MetaAIHackathon
https://t.co/AzpoztARVn
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Why Meta's models: the task needs long-video understanding, small-text reading, object tracking and agentic tool use on the same footage. Muse Spark's 1M context takes a whole 5-minute protocol in one call. @AIatMeta
#MetaAIHackathon
https://t.co/AzpoztARVn
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Lighthouse in a clean Chrome: 100 accessibility, best practices and SEO on both pages, 100 performance on the landing page on desktop. The sample videos now come from the CDN near you.
#MetaAIHackathon
https://t.co/AzpoztAk5P
Thousands of labs own liquid-handling robots that sit idle, because programming them means writing Python, and most scientists don't.
So I built Show, Don't Code for the Meta Global AI Developer Hackathon: film a protocol once, by hand, and get a robot protocol for an Opentrons Flex.
How it works:
π₯ Muse Spark watches the whole video and finds every pipetting step, then looks again at each aspiration moment to tell which tube the liquid came from
π― SAM 3.1 follows the exact tube you click through the recording, and Muse Spark rejects the tracker's jumps
ποΈ Muse Voice Transcribe turns the scientist's narration into evidence
π€ An agent on Muse Spark Open (open weights, on Baseten) runs the robot simulator itself until the volumes add up, then asks you only what it can't verify
π¨ Muse Image redraws each step's own video frame as an illustrated bench card for training the next person, and Muse Spark checks every drawing against its frame
Code generation is deterministic, so model output never becomes executable code.
Then the strongest check: Opentrons' own robot software (robot-server v10.1.0, Flex simulation) runs it end to end. 240 of 240 commands match its own analysis. The same steps also run on Hamilton STARlet and Tecan EVO simulators.
On videos it never saw, it finds 79% of PCR steps, and it tells you plainly where it still needs you.
No video? Paste a published https://t.co/T6dwMtMOsy protocol and get the same robot protocol.
π One file drops the whole run into your lab notebook (eLabFTW, RSpace, Kadi4Mat) as an .eln entry. Every run also ships a reproducibility receipt timestamped in Bitcoin, an RO-Crate on Zenodo, and an MCP server so other agents can use it.
Try it free, no login: https://t.co/8ImCv8jZAB
Full demo (2:53): https://t.co/XGWHmUruPA
Code (open source): https://t.co/0GKLfxt5s4
@AIatMeta #MetaAIHackathon
Last Seen Hashtags on Sotwe
Ψ³ΩΨ―Ψ§ΩΩΨ§Ψͺ
thundr
Seen from Brazil
lesbianincest
Seen from Pakistan
Teen18
fartcommunity ebony
Mombody
Seen from Sweden
ferrarigtc4lusso
Seen from United States
mgai()(()**(****))*video
Seen from Turkey
yenge((((((((*************)))))))) filter:native_video
Seen from Turkey
brosis()****** + filter:native_video
Seen from Turkey
Trends for you
Most Popular Users
Elon Musk 
@elonmusk
241.4M followers
Barack Obama
@barackobama
118.8M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.9M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.4M followers
Kim Kardashian
@kimkardashian
70.8M followers
YouTube
@youtube
68.7M followers
Neymar Jr
@neymarjr
66.4M followers
Bill Gates
@billgates
65.2M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.7M followers