Bhanu K ¯\_(ツ)_/¯

Building technology blocks for Indian healthcare. I'm hiring, DM for details.

Motto: Don't just complain, do something about it.

[Bhanu K ¯\_(ツ)_/¯ @thoughtisdead]

Excited to share the launch of Collab Place, a community learning center.

https://t.co/4HHI72an4B

Promotes learning, collaboration and works to reduce digital divide.

[GPUMonk @GPUmonk]

Sir, I mean no disrespect to Sarvam or to you. As an Indian, I’m genuinely proud that we’re building serious models and infrastructure. That said, I don’t think this is comparable to the DeepSeek moment

Coming to your reply

This is exactly why it’s not an apples-to-apples comparison.
A model optimized for Indic speech, OCR, and local task distribution should outperform general-purpose frontier models on those benchmarks.
That’s specialization, not global superiority.

The confusion starts when tokenizer + domain-specific evaluation is framed as “beating” models designed for cross-domain generalisation

[Burak Eregar @burakeregar]

most people will install clawd and accidentally hand it their entire life

it’s incredible: a 24/7 ai agent on your server that controls your github, calendar, and email via whatsapp/telegram

but stop and think for a second

you just gave an ai autonomous execution rights on your machine and root access to your digital life

if you run this with default settings, you are one prompt injection away from wiping your entire github organization, losing your emails or much worse

before you connect it to anything, you need to lock it down to make sure you and your digital life are secure

here is the non-negotiable security config for clawd: 👇

[Ankur Gupta @getpy]

Operational realities and lessons learnt post shipping AI features

- The ability to switch models at the function or feature level—without restarting your software—should be baked in from day one. ( Also see openrouter )
- Model providers often tweak internal parameters on an ongoing basis that impact throughput, accuracy, latency, and error rate.
- Occasionally, model providers and vendors offer joint discounts that can be leveraged simply by switching models thus saving real costs.
- In scenario where your AI feature adoption is growing and your monthly billing quota wouldn't last, incorporate manual progressive degradation of models. ie. Switch to an older model that offers more tokens per dollar. This works as long as the error rate is acceptable to both the business and product teams. Better than switching off the feature.
- Integrate evaluation set into automated unit tests, allows assessing new models and verifying new prompt or code changes by QA early in the process.
- Teach QA how to build synthetic datasets for input subsets where error rates are high. This takes the load of devs who are doing everything.
- Log and more imp incorporate meta tags in logs against all LLM failures and success. You can use this to optimise prompts via DSPy.
- When launching brand-new features with no prior customer data / input to evaluate against, do a small % rollout and store the customer inputs, output and then use DSPy to optimise your prompt. This cycle will lead to better results next rollout.
- Have tool calling for sensitive or customer profile data ( e.g. PII ). Thus only fetched when explicitly needed and masked. This will also give you an audit trail and you don't do dumb things like injecting it in context window every time.
- Use tools like MLflow and educate your SREs or on-call developers on
- The non-deterministic nature of LLMs, to prevent false alarms. SREs to distinguish between vendor / LLM side issue vs code.
- SREs should know where prompts are stored in the repository.
- Uncapped API quota is a disaster waiting to happen, Include a feature flag from day one to toggle entire AI features off asap if need be.
- Beyond setting aggressive billing alerts, you’ll need a dashboard that provides a granular understanding of which prompts are called -> how many times -> their average cost.
- max_tokens and parallel / batch inference are your friends. Use them.

[Bhanu K ¯\_(ツ)_/¯ @thoughtisdead]

@anuswaram అందరిదీ ఒకటే మ్యూజిక్

[Harsh Nisar @NisarHogaya]

Any 'cracked' AI/MLE engineer who wants to crack government enterprise automation with LLMs and other recent advances in AI?

Any level is fine -> needs to be skilled. Full-time and onsite.

[Bhanu K ¯\_(ツ)_/¯ @thoughtisdead]

@simonw Cogito 70B via ollama, https://t.co/LID8sZMIXE

Structured output

[Hyderabad Python User Group @hydPython]

👋 Say hello to the new face of HydPy!

🎉 We’re excited to unveil our brand new logo - a fresh look that reflects the spirit, growth & energy of the HydPy community.

👏 Huge thanks to @iamaravindsekar & Karuna Solanki for bringing this to life with their amazing design work!

[Eleanor Berger @intellectronica]

🆕 𝗢𝗽𝗲𝗻-𝗦𝗼𝘂𝗿𝗰𝗲 𝗔𝗜 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗖𝗿𝗮𝗳𝘁 👈

➡️ https://t.co/734hMGZdtc

The space for OSS AI software tooling is buzzing with activity⚡ but all we hear about is a few VC-backed closed tools 🫤

I am kicking off a new website to help the community document and learn together.

The first version is simple, but with your help, I hope we can grow this into a great resource.

🫵 YOU CAN HELP:
1. Spread the word - share this post and the link to the website.
2. File issues.
3. Submit PRs.

LFG! 🚀

[Ankur Gupta @getpy]

Notes from completion of a small freelancing project that required near real-time processing of audio files for transcripts. Few hacks that help reduce cost and offer replacement to an expensive SAAS offering

https://t.co/iYcmkRbO8g

[₹₹₹ @vennelacheekati]

The great compute divide

[Ananda Sangeetam @anuswaram]

ఇజం ఇజం ఇజం
నీ ప్రతి ఆలోచన ఒక ఇజం
నీ ఉనికే ఇజం ఇజం
కాస్టిజం, పేట్రియాటిజం
ఇజం లో ఇరుకున్న
నేను అనే ఇండివిడ్యువలిజం
నీ మెదడు ఇజాల్లో నిడిపోయింది
అని అంటే ఆ ఆలోచన మరో ఇజం
రవికిరణం చీల్చగలం
రంగులుగా మార్చగలం అనేది ప్రిజం
రంగులనే చేర్చగలం
రవికిరణంలా చేయగలం అనేది నిజం
నిజం కావాలా నాయనా?
ఇజాలేవి లేనప్పుడు ఉన్నది నిజమే
ఇది నాన్ డ్యువలిజం అనకు
నేను మౌనంగా ఉన్నాను అనుకోగలిగితే
మౌనంగా ఉన్నట్టేనా?
నేను మట్లాడేస్తున్నాను మరి అయితే
మౌనంగా లేనట్టా ?
సరిగ్గా చూడు హాంటెడ్ హౌస్ లో టనెల్ తిరగట్లేదు అనేది నిజమైన ఇజం

[Jarred Sumner @jarredsumner]

In the next version of Bun

`bun init` detects if claude code is installed and adds a "https://t.co/VAvh73IjDt" to tell claude to use bun

[Hyderabad Python User Group @hydPython]

🚀 Get ready for the #HydPy meetup happening on 21 June 2025! From GenAI insights to data engineering in action, catch exciting talks by Vivek Keshore, Sonu Kumar & Sourav Roy.

Don't miss it! RSVP now: https://t.co/MLCm7HN8yL.

Thanks to @AdvanceAuto for generously hosting us!

[₹₹₹ @vennelacheekati]

ఏదో కొంచెం టింకరింగ్ మొదలుపెట్టా....😂😝
https://t.co/ySuB5mmXzi

[Bhanu K ¯\_(ツ)_/¯ @thoughtisdead]

@vennelacheekati ఎవ్రిథింగ్ విల్ బి మోస్ట్లీ బిహైండ్ లాక్డ్ గార్డెన్స్

[Arnav Bansal ⠕ @itsarnavb]

In India 🇮🇳, students in colleges prepare to enter the real world by doing fake things:

- meaningless resume padding projects
- endless interview prep
- placement politics

We can do better. Introducing @quest_sh:

A set of hard but tractable *real world* problems you can solve to prove your mettle

[₹₹₹ @vennelacheekati]

ఈ జాబ్ రిక్వైర్మెంట్ ఉంది. ఎవరైనా తెలిస్తే డీఎం చేయండి

Python Developer
Exp: 5 Years
Location: Vizag/Remote

[khushi @khushhhi_]

if you're going to be in hyderabad this weekend, please dm me or @itsarnavb !

planning to host a small gathering of folks in high school and university to talk about education and deep tech