TEJINDER SANDHU

Building AI that does the work — not just suggests it.
AI engineer for your GitHub repo →

[TEJINDER SANDHU @aurem_live]

Try it on your own small model. It is free. Inspired by MIT CSAIL researchers' Weight-Space Programming blog post (Nov 2, 2026): https://t.co/TEu1izfwyS . Not affiliated with MIT. #LocalLLM

[TEJINDER SANDHU @aurem_live]

76.11 to 85.56. One small AI model. 129,897 numbers changed. No retraining. WSP-Kit (Weight-Space Programming Kit) did it. 5 of 7 real tries failed and shipped nothing. 16 GitHub stars so far. #SmolLM

[TEJINDER SANDHU @aurem_live]

Your one star encourages our team to keep building incredible things — free for everyone — and to grow this community. Send this to one friend who works with AI models. https://t.co/TNcp30BaAg https://t.co/1wNAJ9HUbs

[TEJINDER SANDHU @aurem_live]

Your AI-built app can work and still leak user data.

In Supabase, turn on Row Level Security (RLS) for every table. Without it, anyone with your app's public key may read or change that table.

Then test as a normal user. Is RLS on for all your tables? #VibeCoding #AppSecurity

[TEJINDER SANDHU @aurem_live]

939 clones in the last 14 days. Clones are not people.

Inspired by MIT CSAIL researchers' Weight-Space Programming blog post (Nov 2, 2026): https://t.co/TEu1izeYJk Not peer-reviewed. Not affiliated with MIT. Credit and inspiration: @lateinteraction

[TEJINDER SANDHU @aurem_live]

One tiny AI model: 37.78 to 67.22 on one test. Here is the catch.

WSP-Kit (Weight-Space Programming Kit) changed 0.022% of its numbers. The test counts repeats twice, so the true gain is smaller.

5 of 7 tries failed and shipped nothing. #Qwen #SmallModels

[TEJINDER SANDHU @aurem_live]

Your one star encourages our team to keep building incredible things — free for everyone — and to grow this community.

Try it on your own small model. It is free. Send this to one friend who works with AI models.

https://t.co/TNcp30BaAg https://t.co/1wNAJ9HUbs

[TEJINDER SANDHU @aurem_live]

One tiny AI model: 37.78 to 67.22 on one test. Here is the catch.

WSP-Kit (Weight-Space Programming Kit) changed 0.022% of its numbers. The test counts repeats twice, so the true gain is smaller.

5 of 7 tries failed and shipped nothing. #Qwen #SmallModels

[TEJINDER SANDHU @aurem_live]

A vague bug report gets expensive fast.

"Login is broken" = hours of guessing.

Send 4 things: what you clicked, what you expected, what happened (paste the exact error), and a screenshot.

What do you forget most? #VibeCoding #AIApps

[TEJINDER SANDHU @aurem_live]

Inspired by MIT CSAIL researchers' Weight-Space Programming blog post (Nov 2, 2026): https://t.co/TEu1izfwyS Not peer-reviewed. Not affiliated with MIT. Credit and inspiration: @a1zhang @rikiyatakehi

[TEJINDER SANDHU @aurem_live]

Four questions. Then your small AI model gets a fix.

WSP-Kit (Weight-Space Programming Kit) has an easy mode. Type "wsp auto", answer 4 questions, and it runs by itself.

5 of 7 real tries failed and shipped nothing. That is on purpose. #LocalLLM #ModelEditing

[TEJINDER SANDHU @aurem_live]

Your one star encourages our team to keep building incredible things — free for everyone — and to grow this community.

Try it on your own small model. It is free. Send this to one friend who works with AI models.

https://t.co/TNcp30ACKI https://t.co/1wNAJ9HmlU

[TEJINDER SANDHU @aurem_live]

Inspired by MIT CSAIL researchers' Weight-Space Programming blog post (Nov 2, 2026). Not peer-reviewed. Not affiliated with MIT. https://t.co/TEu1izeYJk

[TEJINDER SANDHU @aurem_live]

Can you fix a small AI model without retraining it?

WSP-Kit (Weight-Space Programming Kit) changes a few numbers inside it. 5 of 7 real tries failed our safety check and shipped nothing.

816 clones in the last 14 days. Clones are not people. #SmallModels #LLM

[TEJINDER SANDHU @aurem_live]

Your one star encourages our team to keep building incredible things — free for everyone — and to grow this community.

Try it on your own small model. It is free. Send this to one friend who works with AI models.

https://t.co/TNcp30BaAg https://t.co/1wNAJ9HUbs

[TEJINDER SANDHU @aurem_live]

Building your app with AI takes a weekend. Fixing it when it breaks can take your whole week.

Tell ORA what's wrong in plain English and it fixes your GitHub code after you approve. Works when your app's code is on GitHub.

Who fixes yours today? #VibeCoding #AIApps

[TEJINDER SANDHU @aurem_live]

See how it works: https://t.co/oY4dXOOVm7

[TEJINDER SANDHU @aurem_live]

Your one star encourages our team to keep building incredible things — free for everyone — and to grow this community. Please share this with a builder you know. https://t.co/TNcp30ACKI #WeightSpaceProgramming

[TEJINDER SANDHU @aurem_live]

Qwen3-0.6B, narrow animals task. Before: 37.78. After: 67.22. We edited 0.022% of its weights (131,072). No retraining. Repeat answers are counted separately by the scorer, so the true gain is a bit smaller. #SmallModels #OpenSource