Rakesh Varudu

Internet Enthusiast

[LNPR Capital @LnprCapital]

Indian CDMOs are entering a major capex cycle.

Stocks to track:

Laurus Labs
Sai Life Sciences
Anthem Biosciences
Piramal Pharma
Syngene
Akums
Blue Jet Healthcare
Neuland Labs
Cohance Lifesciences
Windlas Biotech

Peptides | ADCs | Biologics | Fermentation | OSD

Capex today could shape the next growth cycle.

For research purposes only. Not a recommendation.

#CDMO #Pharma #SmallCap #LNPRCapital

[Sahil Sharma @sahil_vi]

your gut decides what your brain remembers. literally, with a wire.

usc researchers fed rats nutritious food and watched memory circuits fire harder through the vagus nerve, the cable running from gut to brain. sweet water with no nutrition produced no such signal. the study ran this july.

the body writes its memories based on what the meal was worth. junk in, nothing saved.

your grandmother called it eating well. neuroscience just found the wire.

receipts: https://t.co/rJzFxYPMwZ

[abhay jain @abhayjainp]

Growth Triggers for GMM Pfaudler.

Two large orders for peptide manufacturing systems.

Order inflow from newer industries like semiconductors, nuclear, defence, mining.

Edlon uses fluoropolymer fabrication for high purity semiconductor chemical systems.

90% of world’s top chemical companies use Pfaudler glass lined equipment

Vatva can manufacture equipment weighing 350 tonnes & measuring 120mt.

Has 1800 qualified welding procedure across tantalum, zirconium, titanium, Inconel & other critical metals.

Entered nuclear and defence equipment.

Operating leverage is visible in margins.
India glass lined moved from 50 to 60% utilization to full capacity.

[Intrinsic Compounding @soicfinance]

The peptide market: $51 Bn TAM, on its way to $100 Bn by FY34. The pipeline went from ~100 molecules to 800 in development. Semaglutide alone is a $24 Bn drug.

But here's the part that matters throughout the value chain

A peptide is made on an 8-step assembly line: synthesise the amino acids -> anchor them to a bead -> add one link at a time -> cleave -> purify -> freeze-dry -> fill into a pen. Making the peptide is cheap. Anyone can string amino acids together.

The whole moat sits at Step 6: purification. It's slow, it's the hardest capability to build, and it's where the value quietly concentrates.

So who captures it?

-> Upstream (Aminos Acids): commodity building blocks, thin margins but steady.
-> Midstream CDMOs (synthesis + purification): 20-30%+ EBITDA, customer switching takes 2-3 years to re-qualify. This is the sweet spot.
-> Downstream (Dr. Reddy's, Sun, Zydus + Shaily's pen-device near-monopoly in India): highest absolute margins, but generics compress them over time.

The number to internalise: an FDA-inspected peptide purification line with 3+ commercial programs takes a rival 5+ years and $100M+ to replicate.
The largest position is rarely the most durable one. The one that's hardest to copy is.

Disclaimer : Not an investment advice.

[Tijori @Tijori1]

Peptides are emerging as one of the hottest categories in Pharma

Here is a handy list of companies, what they do and their capability depth

[Intrinsic Compounding @soicfinance]

Space Economy value chain in India

Disclaimer - not a recommendation to buy/sell

[Vaidehi @Ai_Vaidehi]

8. Production-level UI component builder:

Think like a senior frontend engineer and build:
• Reusable UI components
• Accessible
• Production-ready

Consider:
• Loading states
• Edge cases
• Responsive design
• Accessibility

Result:
• Component architecture
• Props design
• Implementation
• Usage examples

[Rakesh Varudu @RakeshVarudu]

@LearningEleven Thanks for the quick response.

[Antifragile Thinking @unseenvalue]

Never read outstanding books just once. Return to them again and again; each iteration sharpens your thinking and reshapes how you see the world.

Elon Musk builds what is possible.

Peter Thiel decides what is worth building.

Shane Parrish improves how you think.

Nassim Taleb ensures you survive.

Robert Greene ensures others don’t stop you.

Michael Mauboussin understands how value is created.

Charlie Munger sharpens your judgement.

Howard Marks teaches you how to manage risk.

Adam Grant helps you rethink assumptions.

Daniel Kahneman explains how your mind fails you.

Naval Ravikant clarifies how to build wealth and freedom.

Ray Dalio systematises how the world works.

[Tech with Mak @techNmak]

Sebastian Raschka is one of the most respected researchers in ML/AI education. Period.

And now he's done something quietly brilliant.

He built an LLM Architecture Gallery - a single, browsable reference that maps out the internal architecture of every major open-weight model released in the last few years.

This is a serious research artifact, made free for everyone.

Here's what's inside:

🔹 GPT-2 XL (1.5B)
🔹 Llama 3 (8B)
🔹 OLMo 2 (7B)
🔹 Llama 3.2 (1B)
🔹 Qwen3 (4B, 8B, 32B)
🔹 DeepSeek V3/R1 (671B)
🔹 Kimi K2 (1 Trillion)
🔹 Gemma 3 (4B, 27B, 270M)
🔹 Mistral 3.1 Small (24B) & Mistral Large (673B)
🔹 Llama 4 Maverick (400B)
🔹 Qwen3 235B-A22B & Qwen3 Coder Flash
🔹 SmolLM (1B)
🔹 GPT-OSS (20B, 120B)
🔹 Grok 2.5 (270B)
🔹 GLM-4.5 (355B), GLM-5 (744B), GLM-4.7 (355B)
🔹 MiniMax-M2 (230B) & MiniMax-M2.5
🔹 Kimi Linear (48B-A3B)
🔹 OlMo 3 (7B) & OlMo 3 (32B)
🔹 Nemotron 3 Nano (20B-A3B) & Nemotron 3 Super
🔹 Xiaomi MiMo-V2-Flash (309B)
🔹 Arcee AI Trinity Large (400B)
🔹 Tiny Aya (3.35B)
🔹 Step 3.5 Flash (196B)
🔹 Nanbeige (4.1, 3B)
🔹 Qwen3.5 (997B)
🔹 Ling 2.5 (1T)
🔹 Sarvam (30B, 105B)

And for each model, he links:
→ The original tech report
→ The config[.]json (so you can verify every number yourself)
→ From-scratch implementations where available

But here's what makes it truly special.

He also added short concept explainers, so you're not just staring at boxes and arrows:
→ GQA (Grouped Query Attention)
→ MLA (Multi-head Latent Attention)
→ SWA (Sliding Window Attention)
→ QK-Norm
→ NoPE (No Positional Encoding)
→ Gated DeltaNet

This is the kind of resource that used to require buying 3 textbooks, reading 40 papers, and spending a weekend.

Now it's one link.

If you're studying LLMs, building on top of them, or just trying to understand how the field has evolved, this is a must-bookmark.

[Rimsha Bhardwaj @heyrimsha]

I hope you've found this thread helpful.

Don't forget to bookmark for later.

Follow me @heyrimsha for more.

If you enjoyed reading this post, please support it with like/repost of the post below 👇

[Swapna Kumar Panda @swapnakpanda]

If I were a beginner in System Design, I would start by watching these Videos:

[Katyayani Shukla @aibytekat]

I hope you've found this thread helpful.

Follow me @aibytekat for more.

Like/Repost the quote below if you can:

[Sekhar @LearningEleven]

The Aerospace Ancillaries - Short Stories in one post

1. Dynamatic Technologies

Started delivering complete ship-set of all eight doors for the Airbus A220. If my understanding is right, likes of Sansera and Aequs are working with Dynamatic on this.

Of course, Dynamatic has also started other Airbus programs (not doors) for "A320/A330". If and when Airbus awards doors program for A320/A330 series, Dynamatic could enter into different league! But I have no idea if it would come anytime soon.

But, of course, several other programs for OEMs like Bell, Boeing, Dassault Aviation, Deutsche Aircraft, and Thales. Additionally, Dynamatic has also developed 4-5 versions of Drones which could turn out to be a good optionality. But of course, consolidated earnings growth would depend on how they deal with their other two verticals.

They do have couple of programs (including 220 doors) that could generate > 400 cr each (9MFY26 aerospace revenue was 565 Cr), so, opportunity is big! Only question is can they execute well and capitalize!?

Any correction towards 8000-8500 could be a decent entry point

2. Aequs Ltd

They have two verticals. Aerospace vertical is doing very well but the Consumer vertical is their achilles heel. It is not only loss making but hurting the overall profitability of the company severely. The aerospace vertical perhaps is best in class with a vertically integrated setup (hardly 2 or 3 in the country with such capabilities). Their consumer vertical has two sub segments, Toys and Electronics.

Toys division isn't doing much although they commenced deliveries to Mattel and that should help. But their Electronics sub-segment could be the wildcard here. They have started supplying to (supposedly) Apple and this one is expected to pick up momentum in next 2-4 quarters and thereby driving the overall profitability significantly upwards! Some analysts believe that Consumer Electronics could become bigger than Aerospace vertical and if that happens, you are looking at a big show!

Rs 100-115 could be a decent entry point but one would need to wait for 3-5 quarters to see significant earnings jump.

3. Sansera Engineering

Primarily still an auto ancillary but transformation towards better product mix has started. If all goes well, they are expected to post 1000 Cr in FY28 in ADS segment (Aerospace, Defence and Semiconductors).

Considering that their H1FY26 ADS revenue was 86.4cr, you could see why market was excited to re-rate it already. But honestly, a good chunk of this optimism is now captured in this. Any correction towards 1700-1800 would be a good buy. Good management and better capabilities.

4. Azad Engineering

Just like Dynamatic, shifting to Aerospace more and more. They manufacture one of the best and most complex products with huge moat. They have also undertaken huge capex and they probably have few execution challenges. If and when they commission the full capex, we might be looking at a giant! Valuations aren't cheap at all. This one takes good support around Rs 1400

5. Unimech

They supply tools for MRO and that's a cyclical/lumpy business. Unless this business is available at very attractive price, this wouldn't be my top priority. There has also been a rumor that they lost a major client, but on that note, they are trying to expand into other verticals and geographies.

6. Rossell Techsys

They produce Electronic Wiring components and that's a very critical component in aircrafts. However, most of their Aerospace work is limited to defence aircraft and they don't have much presence in commercial aerospace which is actually the sweet spot.

But they have started supplying electrical wiring and harness to semiconductor and space companies and that could be a game changer if they can scale it up. But valuations are super rich.

7. Raymond

Just like Sansera, primarily an auto ancillary company but they have some solid expertise in Aerospace vertical. They are setting up a large plant at Sri City in AP but first leg of that commercialization is about consolidating multiple manufacturing centers they have in Karnataka.

Hence, you may not really see significant incremental capacities at least for next 6-8 quarters. But they are doing 100s of FAI (First Article Inspection) and depending on how many will be successful, might determine the future opportunities.

8. Sigma Advanced

They are originally a defence firm and then acquired 45-47% in Indrajaal which is anti-drone specialist (deployed at western command center). They have recently acquired Nasmyth group (UK) and that group supplies to Tier-1 aerospace firms. So, in a way, Sigma now has a sizable aerospace vertical (annual revenues of 700+ Cr). How they can grow it further is the key. If someone has 3 years sort of view, this could turn out to be a giant player if they scale up all three verticals!

However, the original firm Megasoft which acquired Sigma and adopted the name has had very bad history!

Disc: This is not a comprehensive list and none of these are buy or sell recommendations.

[Rakesh Varudu @RakeshVarudu]

@DAIEvolutionHub ML

[Tech with Mak @techNmak]

🚨 This is the best way to learn how LLMs work.

Interactive. 3D. Step-by-step.

Covers:
→ Embedding
→ Layer Norm
→ Self-Attention
→ MLP
→ Transformer layers
→ Softmax
→ Output

Stop reading papers. Start seeing.

Link in comments.

Save this immediately.

[Siraj Raval @sirajraval]

I built a 3-agent AI that taught me calculus in 5 minutes.

One agent listens to my reasoning.
One draws on a whiteboard.
One talks me through it.

It doesn't give you answers, it teaches you how to think.

Open source. Free on my GitHub.

[Michael Pyrcz🌻 @GeostatsGuy]

I really love teaching Bayesian linear regression. It’s the perfect introduction to #Bayesian methods for #MachineLearning.

Spoiler alert: they’re awesome. 🔥

In class, I walk students step-by-step through:

1️⃣ How the prior is updated with the likelihood to produce the posterior
2️⃣ How we sample that posterior using Markov Chain #MonteCarlo (MCMC) — specifically Metropolis sampling

Then we open up my interactive #Python dashboard and actually do it:

Sample the Markov chain of model parameters, compute the acceptance probability directly from
prior × likelihood ∝ posterior

Theory → algorithm → live visualization.
No black boxes. Just probability in motion. 🚀

That moment when students see the posterior emerge from the sampling process? Completely stoked. 🤘

I share the full interactive notebook here:
https://t.co/K4L8GjQwOU

#GitHub

[Nav Toor @heynavtoor]

🚨 BREAKING: Someone just made 70B parameter models run on a single 4GB GPU.

It's called AirLLM. No quantization. No distillation. No pruning. Just raw 70B inference on hardware that costs less than a dinner.

You can even run Llama 3.1 405B on 8GB VRAM.

Here's how it works:

→ Decomposes the model layer-by-layer
→ Loads only one layer into GPU memory at a time
→ Runs inference, moves to the next layer
→ Prefetches the next layer while computing the current one
→ Supports 4-bit and 8-bit compression for 3x speed boost

No cloud API. No $10K GPU. Just pip install airllm and go.

Here's the wildest part:

It supports almost every major model — Llama, Qwen, Mistral, ChatGLM, Baichuan, InternLM — and it auto-detects the model type. One line of code to load. One line to generate.

Works on Linux, macOS (Apple Silicon), and even Google Colab free tier.

Your old gaming laptop can now run the same models that needed an A100.

100% Open Source. Apache 2.0 License.