Any tooling to get a breakdown of token costs for LLMs ? Did a trial run of analysing a Whatsapp conversation using Claude APIs, and it was pretty darned expensive for just one result. Now trying to figure whether the data was so much, or whether there's meta-data bloat

Home / office space for rent in Indiranagar very near the metro station at a 2 minute walk. Posting because it's always nice to have friendly faces around. Not affiliated to the property in any way. RT if you know anyone who might be interested

Just measured the TDS in the water. Distilled water: 6 ppm Tap water : 339 ppm RO water : 17ppm Pitcher water : 19ppm Do the readings look reasonable?

What are things you wish you could do on your TV that you find are impossible / frustrating / difficult / time-consuming etc to do today ?

Video calling ✅ Onboarding flow ✅ Invitation mechanism for family / friends 🔄

Anybody using local models for code completion ? Planning to run one on my laptop to test out performance (upto 16GB model size is ok). I don't see any "coder" model releases newer than 8 months ago

Calling from mobile phone to TV works well, between all combinations of devices ! Got the first calls from Mumbai to Bangalore in. Bonus : was able to check in to my living room while out 😁 The kid is quite excited about this, and starts dancing in front of the camera now !

Fun Saturday building a rack for my home server at Rootconf , Kiran Jonnalagadda 🦋 @jackerhack.ing has put together a fantastic guide on how to get started here jackerhack.ing/notes/20250312… and Gaurav Chaturvedi is experimenting with customisations for his not-so-mini home server 😄

There's probably a new market opening up soon for consultants / agencies "We help you fix your AI written code so that you can scale beyond 100 customers without continuing to burn through money". Let's see how terrible (or not) a take this is

All visits to hardware stores are done (🤞). Everything has been acquired, should be able to complete the assembly tonight

This passive heat sink seems to have helped drop the core temperature from 85° to 65° cutting out throttling and stuttering, imagine what a fan can help achieve ! Now onwards to endurance tests and further development

Rack for the home server setup is ready and mounted. Time to check for stability (literally also, if you noticed the jugaad for the base ;) ) !

Hey Zerodha I tried using your MCP server on AnythingLLM Desktop on Linux and it doesn't seem to work, getting the below error. Can you help me get it up and running ?

Turns out I'll have to compile a kernel now since *that specific version* of the kernel for that *that specific board* has *that specific flag* disabled which uses the GPU

It is equal parts enjoyable and maddening that you are passionate about two completely disparate problem areas. Balance looks more like swinging between both extremes repeatedly 🤷‍♂️

AI is accelerating everything. However, some things are timeless, like sheer lack of testing. HDFC Bank asks for a 6 digit OTP during login but doesn't accept it when entered. Can't go back to old interface, and can't login. ReserveBankOfIndia has its hands full

Small town life is so alluring, partly because we have nostalgia of some of our own childhood, and it is such a relief from the stresses of everyday big city life

To fix an error, Clause sonnet with Cursor recommends "We are using require which is commonjs, let's use import which is ESM". I accept. New error, next reply. "There is a missing dependency, add 'require' statement." 🤦‍♂️ Conclusion: Ghajini has better memory than LLMs.

Given that B2B AI needs to be fed a ton of "context" around your company and data for it to *start* working well, how soon before you let a less-than-scrupulous "AI tool" in that exfiltrates your data accidentally, or worse, maliciously ? How do you prevent that ?

What stops me from building even more and even faster? I let the agent run and switch tabs to either lookup something else or watch a video and then switch back to that tab many minutes later instead of seconds later.