5/ Still hard:
- Asian + Arabic metadata extraction (regex, mediocre)
- ECHA blocked by Cloudflare WAF
- No translated text stored
Plugin source MIT-0. DM if you read CN/AR text.
https://t.co/VlhiMMskBm
🧵 How I built a regulatory data plugin that doesn't hallucinate.
Most "AI for legal" tools either (a) make up article numbers, or (b) only cover EU+US then fall off a cliff.
So I built the boring infra. Here's how ↓
4/ Quality gate at every insert:
- min 200 chars
- legal markers present
- no HTML noise
- content_hash for change detection
A not_found row beats a garbage resolved row. Garbage in embeddings = garbage agent answers.
Everything's open source — code, prompts, per-platform doctrines.
Fork it. Run it on your brand. Break it and tell me what sucks.
→ https://t.co/SPm4l6d7wq
Everyone's building AI chatbots for websites.
I built a network of AI agents that lives inside Instagram, WhatsApp, Reddit, TikTok, X and Facebook.
24/7. Shared memory. One voice across all 6.
Just open-sourced the whole thing 🧵
Memory is shared across all 6.
A convo that starts in an IG DM in March can pick up in WhatsApp in May — the agent remembers.
That's the part most "AI agent" startups skip.
Day and night, the Factory has been working relentlessly to print out the posters
The propaganda is on its way, see you in the streets of Paris in a few days…
Catch us at @nft_paris February 24/25th and take part in our mix-reality experience turning you into a New Face avatar
To touch tomorrow's mobility, you have the choice :
1. come to our stand at @mondialdelauto,
2. dive into the immersive #Mobiloop experience,
3. and why not both?
Get ready, it starts on October 17th, in hall 6 of the ParisMotorShow…#TomorrowStartsNow#WeAreMobilizers