Ayush Khati

Backend Engineer
Building scalable microservices & high-availability systems
2.5K+ users | APIs • Distributed Systems • AI

[Ayush Khati @Dropclause2]

@alexabelonix Right can't wait to test it

[Ayush Khati @Dropclause2]

Claude Fable 5 and Mythos 5 are out.

More interested in:

• agent reliability
• tool-use robustness
• hallucination rates
• adversarial resistance
• long-context performance

New models are exciting.
New failure modes are even more interesting.

#AI #LLM #Claude #AIAgents

[Ayush Khati @Dropclause2]

VibeSec v0.4.0

Now scans public GitHub repos directly:
vibesec scan https://t.co/6rnhos4NC1

No clone needed. Finds real vulnerabilities
in seconds.

pip install --upgrade vibesec
https://t.co/3j88Tt7NEi

#CyberSecurity #OpenSource #Python #DevTools #AppSec #OWASP #GitHub #AI

[Ayush Khati @Dropclause2]

Built a security scanner for AI-generated code.

Ran it on my own projects.
Found 4 vulnerabilities in my own code.

45% of AI-generated code has critical security issues.

pip install vibesec
https://t.co/3j88Tt7NEi
#VibeSec #Security #OpenSource #Python #CyberSecurity #AI

[Ayush Khati @Dropclause2]

@Jia_builder AI/ML here check my profile

[Ayush Khati @Dropclause2]

Would love feedback from the AI community on the
scoring approach — especially the hallucination
detection methodology.

GitHub: https://t.co/tB0twWNP56

cc @karpathy @GroqInc

[Ayush Khati @Dropclause2]

Built an open-source LLM eval framework 🧵

→ 27-test suite: hallucination, safety, adversarial
→ 3-tier judge chain scoring
→ 86% accuracy vs 50% random baseline
→ Regression tracking across model versions
→ Free to self-host

https://t.co/u70gMmnt5D

#LLM #AI #OpenSource

[Nandini @N_and_ni]

I have ranked tech influencers …. Do you guys agree ?

[Ayush Khati @Dropclause2]

@md_kasif_uddin Money and Curiosity

[Ayush Khati @Dropclause2]

@mchulet Let's build together
https://t.co/3WzLp9yCQ5

[Ayush Khati @Dropclause2]

@sickdotdev Improving my Physics-Informed Neural Networks – PDE Solver
https://t.co/3WzLp9yCQ5

[Ayush Khati @Dropclause2]

@GoogleAIStudio Modifying my Physics-Informed Neural Networks – PDE Solver needs some changes in logic code

[Ayush Khati @Dropclause2]

@Adidotdev I’d pick based on leverage:
Claude if the problem is unclear,
Codex if the task is defined.
Tools don’t matter—workflow does.

[Ayush Khati @Dropclause2]

@Saanvi_dhillon As for me I started with MySQL
So, for me MySql is beginner friendly. But, it just depends what you start with

[Ayush Khati @Dropclause2]

@hackernoon Agreed. Prompts are just one layer.
The real work is in orchestration, data flow, and making outputs consistent enough to trust in production.

[Ayush Khati @Dropclause2]

@md_kasif_uddin We'll I am waiting for Claude Mythos

[Ayush Khati @Dropclause2]

Anthropic’s Claude Mythos isn’t public—and that’s the signal.
AI that can find and exploit vulnerabilities changes everything.
Security is no longer optional.

#AI #CyberSecurity #Tech #Developers #InfoSec #AITrends #Programming #TechTwitter #claude

[Ayush Khati @Dropclause2]

@anjalinirwal02 I'm here

[Ayush Khati @Dropclause2]

🚨 This repo hit 50K stars in 2 hours… then 100K+.
Why?
Anthropic accidentally leaked the code behind its AI coding agent.
Within hours:
→ devs analyzed it
→ rebuilt it
→ open-sourced alternatives
But this isn’t the real story.
The leak exposed how modern AI agents works

#AI

[Ayush Khati @Dropclause2]

I built an interactive Physics-Informed Neural Network (PINN) simulator that learns to solve the heat equation using deep learning instead of traditional numerical solvers.
Live demo:
https://t.co/lYiWWIIdyM

#MachineLearning #AI #Physics #PyTorch #DataScience #softwareengineer