Rakesh Dhote, Ph.D.

AI & ML Specialist | Model Risk, Validation & Governance | Open-Source Builder | 2× Triathlete

[Rakesh Dhote, Ph.D. @rakeshdhote]

@dr_cintas That’s pretty wild, I didn't know you could extend context windows like that!

[Rakesh Dhote, Ph.D. @rakeshdhote]

@DaveJ That's a neat trick for getting LLMs to help with app documentation and debugging.

[Rakesh Dhote, Ph.D. @rakeshdhote]

@RoundtableSpace I'm curious how well Claude can handle really convoluted, legacy codebases with that approach, honestly.

[Rakesh Dhote, Ph.D. @rakeshdhote]

@_lewtun Wow, that's pretty wild, gonna have to try that out!

[Rakesh Dhote, Ph.D. @rakeshdhote]

@_lewtun That’s wild, local models keep getting good enough that “researcher on your laptop” doesn’t sound like a joke anymore, it sounds like a weekend project 😅

[Rakesh Dhote, Ph.D. @rakeshdhote]

Most AI tools need Docker, 16 cores, and a cloud account.

This one ships as a 20MB binary.

✦ Karpathy LLM wiki built-in
✦ Agentic AI with tool use
✦ Workflow + scheduler
✦ Discord / Slack / Telegram channels
✦ Run as Daemon / Desktop / CLI / TUI / REST API
✦ Plugin system (any language)
✦ Shared AI brain across all surfaces

Open source. Single executable. No bloat.
https://t.co/dTfApwuMaK

#OpenSource #LocalAI #RustLang #DevTools #AIAgent #SelfHosted #BuildInPublic

[Rakesh Dhote, Ph.D. @rakeshdhote]

Came across this lightweight tray tool for AI agents:
• monitors multiple agents
• alerts when input is needed
• stays out of the way
Feels like early-stage but genuinely useful.
https://t.co/UJhPDavRe4

#AIagents #StartupTools #OpenSource #Productivity
#OpenSource #AItools

[Rakesh Dhote, Ph.D. @rakeshdhote]

@Hesamation Pretty wild, but also a bit eerie how much of job hunting is becoming automation. Curious how well it handles edge cases and whether companies will start filtering these out too

[Rakesh Dhote, Ph.D. @rakeshdhote]

@om_patel5 Most of the “Claude is expensive” take is just self own in disguise. If you’re loading dead tools, rereading the same files, and letting cache expire constantly, the model isn’t the problem, your workflow is.
Thanks for sharing.

[Om Patel @om_patel5]

THIS GUY AUDITED 926 CLAUDE CODE SESSIONS AND FOUND MOST OF THE TOKEN WASTE WAS ON HIS SIDE

everyone is blaming anthropic for the limits, so he decided to actually look at the data

858 sessions, 18,903 turns, and $1,619 estimated spend across 33 days

here's what he found:

1\ one default setting was burning 14,000 tokens per turn

Claude Code loads the full JSON schema for every tool into context at session start. whether you use them or not. 20,000 tokens of tool definitions sitting there on every single turn.

the fix: one line in your settings.json

"ENABLE_TOOL_SEARCH": "true"

context dropped from 45K to 20K instantly. across 858 sessions that one setting was wasting an estimated 264 million tokens

2\ cache expiry is the single biggest waste

54% of his turns came after a 5+ minute idle gap.

every one of those turns re-processed the entire conversation at full price which caused a 10x cost jump

you go grab coffee. come back 5 minutes later. type your next message. everything rebuilds from scratch. the context didn't change. you didn't change. the cache just expired.

12.3 million tokens wasted on idle gaps alone

3\ 42 skills loaded. 19 of them used twice or less across 858 sessions.

every one of those skill schemas sat in context on every turn eating tokens for nothing.

4\ 1,122 redundant file reads where the same file was read 3+ times

one session read the same file 33 times.

he ALSO built a full token auditor dashboard that shows you exactly where your waste is coming from

19 charts, opens in your browser, free AND open source

[Garry Tan @garrytan]

My Karpathy-style git wiki knowledge base for OpenClaw got to 2.3GB and I know git limit is 5GB so my GStack autoplan skill one line prompted this spec for my upgraded GBrain with SqlLite.

This will be MIT license open source soon.

https://t.co/ZJmyJxQpt8

[Rakesh Dhote, Ph.D. @rakeshdhote]

@garrytan That sounds wild in the best way, 2.3GB for a knowledge base is no joke. Excited to see the open source drop, especially with the SQLite upgrade 👀

[Rakesh Dhote, Ph.D. @rakeshdhote]

@aigleeson That’s wild, but also kinda exciting. If it really verifies every claim against sources, that could save so much time for anyone doing research.

[Rakesh Dhote, Ph.D. @rakeshdhote]

@NickSpisak_ @karpathy @steipete @tobi This is pretty wild in a good way. Feels like a solid brain dump system for people who actually want their notes and data to talk to each other.

[Rakesh Dhote, Ph.D. @rakeshdhote]

@zhengyaojiang Classic HPO still wins when you need control and repeatability. Autoresearch looks better on paper because it’s optimizing the whole pipeline, not just knobs, so yeah it can look faster and cleaner across the board.

[Rakesh Dhote, Ph.D. @rakeshdhote]

@sharbel The zero-shot forecasting aspect is particularly impressive, seems like it could save a lot of resources.

[Rakesh Dhote, Ph.D. @rakeshdhote]

@heyrimsha This sounds super useful! Having something that constantly monitors for changes is a real time-saver. Thanks for sharing!

[Rakesh Dhote, Ph.D. @rakeshdhote]

@_avichawla Great for agent safety! The proxy layer sounds like a crucial step for anyone looking to integrate real-world data into their agent workflows.