Harsh

AI/ML | Data Science | Web3 | Crypto

[Harsh @Harshcodes1]

@alokbishoyi97 hellooo add me

[Kunal Shah @kunalb11]

What we call talent is often just the combination of:

A deep need to win and high agency
The ability to learn fast from mistakes
A beginner’s mind that never disappears

The common thread: an unusually high rate of learning.

[Mira Murati @miramurati]

Today we're sharing our work on interaction models. A new class of model trained from scratch to handle real-time interaction natively, instead of gluing it onto a turn-based one.
https://t.co/MoS5s4cm60

[Harsh @Harshcodes1]

@ashwathama U should share your journey , how u got through all of this i have just started in ml-dl space working for a startup will be inspiration for all of us out here , best wishes

[Harsh @Harshcodes1]

@luthiraabeykoon @karpathy Crazy build

[luthira @luthiraabeykoon]

We implemented @karpathy 's MicroGPT fully on FPGA fabric.

No GPU.
No PyTorch.
No CPU inference loop.

Just a transformer burned into hardware, generating 50,000+ tokens/sec.

The model is small, but the idea is not: inference does not have to live only in software 👇

[Harsh @Harshcodes1]

@YashviDhruvv Great on the same track will make it for sure

[aditya @adxtyahq]

execution beats intelligence every single time

[Harsh @Harshcodes1]

@SurajJ326311 lets connect

[Harsh @Harshcodes1]

@suraj_sharma14 build ship repeat is the only way out

[Suraj Sharma @suraj_sharma14]

AI engineers are printing money right now.
But only if they know this:

Most people are learning the wrong things.
Courses won’t get you hired.
Skills will.

Here’s what actually pays in 2026:

→ Building end-to-end LLM systems
(not just calling APIs)

→ Working with real data pipelines
(cleaning, chunking, retrieval)

→ RAG that actually works in production
(not tutorial-level demos)

→ Inference optimization
(vLLM, batching, caching)

→ Evaluations
(DeepEval, human feedback loops)

→ Agents (only where needed)
(LangGraph > hype wrappers)

What companies actually want:

→ Someone who can ship
(not just experiment)

→ Someone who understands tradeoffs
(latency vs cost vs quality)

→ Someone who can debug broken outputs
(not blame the model)

→ Someone who thinks in systems
(not prompts)

The gap is simple:

Most people are learning tools.
Few people are building systems.

That’s where the money is.

If you’re learning AI right now:

Stop collecting certificates.
Start shipping projects.

[Harsh @Harshcodes1]

@ravikiran_dev7 True biggest bottle neck is inference

[Harsh @Harshcodes1]

Everyone understands model training.
Very few understand inference.
But inference is the real bottleneck:
• Runs continuously
• Scales with users
• Drives most production cost

I’ll be breaking it down step by step over the next 2 weeks. 🧵

[Harsh @Harshcodes1]

@ananyashaswat @X hey lets connect

[Harsh @Harshcodes1]

@miless_15 ping me if u get a good one Im trying to search for one too

[Harsh @Harshcodes1]

@KiranKS hey do you come daily for a run on the flyover i go through this route everyday would be great if we meet

[Harsh @Harshcodes1]

@3rdEyeDude I Go to run everyday on jaidev flyover one side is unopened for vehicles i see everyday atleast one accident everyone is overspeeding

[Harsh @Harshcodes1]

@lilbrokentwobb @BangaloreRoomi Unable to dm you please can you dm

[Harsh @Harshcodes1]

@wojakcodes Good

[Harsh @Harshcodes1]

@317O00O @FlatsnFlatmates @BangaloreRoomi @GruhamBot Hi intrested can u dm me more details unable to dm you