Vinay Kumar

Building Gaussian - AI agent Runtime security for Developer workstations | Founder & CEO - @Sudoviz

[Vinay Kumar @vinkrp]

Check out the latest article in my newsletter: The model isn’t the risk. The workstation is. https://t.co/ZKEdCPK42m via @LinkedIn

[Vinay Kumar @vinkrp]

PROMPT INJECTION IS A FREE BUG BOUNTY PROGRAM FOR YOUR AI AGENTS

[Vinay Kumar @vinkrp]

When it comes to AI - 'rm -rf /' is the most feared

[Claude @claudeai]

Introducing Claude Code Security, now in limited research preview.

It scans codebases for vulnerabilities and suggests targeted software patches for human review, allowing teams to find and fix issues that traditional tools often miss.

Learn more: https://t.co/n4SZ9EIklG

[Vinay Kumar @vinkrp]

💸 $15,000 · ⏳ 7 days · 🤖 1 AI ➕ 🥇 YOUR IDEA

Here's the experiment:

I'm handing Codex + TuringMind a $15K cloud + API budget and 7 days to ship a real product.

From architecture to deployment — fully autonomous.

But here's the twist: 💪 you decide what gets built.

Drop your craziest product idea below.

The winning pitch gets:
→ Built live, documented publicly
→ 1% equity in the resulting product
→ Credit as co-creator

The catch?

The AI has to follow TuringMind's engineering rails — specs, tests, verification gates.

No vibe coding allowed.

[Vinay Kumar @vinkrp]

Only Pain Point left right now ....

[Vinay Kumar @vinkrp]

Real Impact of Anthropic / Aardvark on Code Security :)

[Vinay Kumar @vinkrp]

We need a better code reviewer in the Age of LLM generated code. Meet TuringMind AI - A Code reviewer that understands your entire codebase, not just the diff.

https://t.co/Nx6F0wdSht

#turingmindai

[Vinay Kumar @vinkrp]

Day 2/30 of Building in Public : I let my AI roast its own source code.

I just unlocked the shortest feedback loop in software engineering.

Normal product cycle: Build → Ship → Wait a week → Get feedback.

My cycle: Build → Review → Fix

Without memory, I’d have to prompt it every time to ignore that file. With memory, it learned the context once and moved on.

Follow along for Day 3 -

https://t.co/07evjUVshi

[Vinay Kumar @vinkrp]

AppSec is hard because we've spent a decade telling developers to "think like hackers" when their job is to "act like builders."

We don't need developers to be security experts; we need security experts to build better developer tools that make it impossible (or very difficult) to do the wrong thing

[Vinay Kumar @vinkrp]

Classic AppSec has decades of infrastructure. AI has raw capability but zero plumbing.

Maybe one eats the other.
Maybe they merge.
Maybe I'm asking the wrong question.

So I'm starting with tmind — a Claude Code Skill + Memory for the important stuff

https://t.co/RftbAoTfJv

[Vinay Kumar @vinkrp]

Starting 2026 with an experiment

Day 1 of building AI Code Reviewer infrastructure in public - no idea if this works.

What I’ve learned so far 👇 Claude catches things SAST tools miss. The stuff that actually matters.

#ClaudeCode #SAST #CodeReview

[Vinay Kumar @vinkrp]

But teams going all-in on AI review keep hitting the same walls:

🎯 Claude forgets everything between sessions
🎯 Same false positives, every single day
🎯 No audit trail (good luck with compliance)
🎯 Scaling gets expensive, fast

AI has raw capability — but no plumbing.

[Vinay Kumar @vinkrp]

Claude Code Skill to perform deep code reviews locally

https://t.co/CbqMLe3pJh

[Greg Brockman @gdb]

Introducing Aardvark, our agentic security researcher:

[Andrej Karpathy @karpathy]

Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI.

It weighs ~8,000 lines of imo quite clean code to:

- Train the tokenizer using a new Rust implementation
- Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics
- Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use.
- SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval)
- RL the model optionally on GSM8K with "GRPO"
- Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI.
- Write a single markdown report card, summarizing and gamifying the whole thing.

Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc.

My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved.

Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

[Greg Brockman @gdb]

hard to beat flow state at midnight while working on a worthy problem

[Vinay Kumar @vinkrp]

Check out the latest article in my newsletter: 🎉 Launching Driftcop – an open-source SAST Scanner for AI agent tooling! https://t.co/w9FFrQ5PK8 via @LinkedIn