Munish pathania

AI/ML
Ex-Core Team & ML Engineer

[Munish pathania @munishtwts]

@idkwhyvi62159 No, that this type of art existed and is made on nails.

[Munish pathania @munishtwts]

@idkwhyvi62159 Viral content hai yae toh
honestly I didn't even know about it, these things also exist 😐

[Munish pathania @munishtwts]

@idkwhyvi62159 🫡

[Munish pathania @munishtwts]

@shyam_tawli will I get a referral?? 😄

[Munish pathania @munishtwts]

@sur_singla Cool build.
Open source it will love to contribute

[Munish pathania @munishtwts]

@rambuilds_ Yeah counting finger's and posting it in linkedin and they thought they have done it 🤣

[Munish pathania @munishtwts]

@247pmstudio

[Munish pathania @munishtwts]

Built around 953K+ unique Hindi passages, multilingual embeddings, FAISS retrieval, semantic chunking, structured agent orchestration, grounding checks, query logs, latency evaluation.

Not just prompt → answer. actual end-to-end RAG pipeline.

https://t.co/3Npvlj7D54

#RAGInGoa

[Munish pathania @munishtwts]

@SakshiSugandhi Because they don't know what to prompt for that, and eventually, there are a lot of tutorials to build a Next.js app.

[Munish pathania @munishtwts]

Bruh 🤯

[Danis @DanisMCFC]

Pep Guardiola gets emotional after 2-0 defeat against Juventus and tells the team about his divorce with his wife.

“In your life, whatever you’re going to do, do it with passion.”

[Munish pathania @munishtwts]

@sigmoidfunc_ Just wait for the clickbait courses to erupt in the market.
"become inference engineer in 1 month" 😅

[Munish pathania @munishtwts]

@WokeFuneral they don’t have answers to our questions + they known our protests is peaceful. so they are doing what they do best "Ignore and don’t answer"

[Munish pathania @munishtwts]

@harshithio @mem0ai peak 😅
solving problem vs problem solved.

[Munish pathania @munishtwts]

@shyam_tawli @grok @archiexzzz Start with OS in depth to NNs,DL then about systems.....

[Munish pathania @munishtwts]

True as hell

[Ayussh Sanghi @ayusshsanghi]

Gen Zs thank your parents; otherwise, you won't have a roof to live under.

The real estate is in a bubble, and it will never burst.

It is impossible to purchase one now.

[Munish pathania @munishtwts]

@_khusheyyy 100+ jobs 100+ rejection event loop

[Munish pathania @munishtwts]

@ThedatagGuy should build a micro-version of this for single multi-core CPU with python queue and simulate this.

[Munish pathania @munishtwts]

Yes exactly voice prompt and context is becoming the new way to feed LLMs but it lacks the context engineering we can do with text based.

Solving this will be next big thing and will make AI more mainstream.

[manavv @manavvnotop]

everyone is hating and asking how a "simple speech-to-text app" raises $280M. i reverse engineered the @WisprFlow a while back, purely to understand how they built it. i kept it to myself then.

here is what i learned:

> the gatekeeper is a tiny model:
before anything expensive runs, a small, fast model decides what you actually said: a message to send, a command to run, an edit to make. most of your speech never touches the big models.

> one problem, multiple models:
they don't pick one model. every job gets a different one: formatting (one is even fine tuned specifically for emails is my guess), voice commands, rewrites, live meeting notes. even the ASR side is not a single model. they run their custom models. not just off-the-shelf Whisper or NVIDIA Parakeet model.

> dictation is a pipeline, not a model call:
voice does not go mic -> text. it goes raw transcription -> formatting -> context rules -> matched to your writing style. the product is much about post-processing as it is about speech recognition.

to be fair: this is all from reverse engineering. so i could be wrong about parts of it. but, this is just what i found.

the ASR is the easy 90%. the routing, model selection and personalisation is where the real depth is.

[Pranav Mailarpawar @pranvtwt]

Thanks a lot for making the RESUME SCREENING process even more unbearable for everyone.

I genuinely don’t understand this obsession with “making job applications easier.” LET IT BE HARD. Let people suffer a little.

First, the Telegram vaale Bhaiya/Didi turned every job application into a spam festival, and now these absolute morons are sweeping up whatever tiny shred of hope was left.

At this point, just hand everyone a job and shut the internet down.