Ankita Tripathi

People start with 3 LPA and reach 30. People start with sharing a room in an apartment to living alone in a 2bhk. People start with nominal English sentences to being fluent. People start with being camera shy to being great creators. People start with ordering food online with coupons from various sites and then.. well that doesn’t change. But see how easy it is to think when it is a journey and not constantly running behind a FAANG tag or 50LPA job. The important part is starting. 30 secs plank to 2 mins. 2kgs to 20. Start today.

@MaziyarPanahi Powerful. Going to try it.

Medical AI models now run on iPhone. No cloud. No API. OpenMed 1.0.0 just shipped. MLX backend for Apple Silicon. Swift package for macOS and iOS. 200+ PII detection models across 8 languages. pip install openmed Open source. Apache 2.0.

Name a movie you've seen more than 7 times with just a GIF

India was late to hi-tech manufacturing. Now "Physical AI" is fast-forwarding legacy plants into smart manufacturing. Some interesting use cases: @DeltaElectIndia: collaborative robots & predictive systems cutting downtime. @perceptyne: AI dexterous robots for assembly that adapt like humans. @SyrmaTechnology: pushing hybrid human-robot environments. Legacy + smart tools = globally competitive manufacturing. This blend is the real next gear for Make in India.

Signed up for my first Hyrox in the pro category and it was all worth it! Regretted the decision till the very moment I started and after that it was such a breeze. The people who signed up for pro were mostly the teams from US, UK, and Japan and I realised how crazily quick you have to be to be at par with them. May be next time!

tried the beetroot parantha with sweet potato halwa (sugarless). Turns out to be epic 😀

Interesting to see that the govt of the most populous state went ahead with Puch AI and the biggest one that every person had to suggest Sarvam as an alternative. In a population as large as India, we don’t even have at least 3-4 good alternatives to take this initiative ahead. Just 1.

@arpit_bhayani wondering why are women not interested to learn System Design.

The System Design Feb Cohort ended today, and I will remember it for how number-heavy we went in all the systems we discussed, and all the fun chats we had before and after sessions :) The S3 design discussion we had is exactly how it is implemented in the real world. The Cricbuzz design was verbatim how it runs in prod - that is how practical the course has been. Getting this kind of acknowledgment from people is priceless. One also mentioned how he discussed the system with his friends, and the kind of questions he asked were the same as those interviewers asked them. Super happy, again, to help people rewire their brains to think in a highly practical sense, with code and all. In every cohort, I add something new. In this one, I added practical aspects around moving data from Redis to DynamoDB and how to actually do it with minimal intervention. It's so fun to see people getting their spark back and, more importantly, a practical understanding. While setting the context, I laid out the design where most people would stop, thinking that was it. Then, when we moved into the implementation nuances that required re-architecting, that is where they had the aha moment. Teaching is not about stating facts, but about helping people build the right mindset through mental models and frameworks - and that is what I do through this cohort. It is not just ~30 systems we discuss, but being ready to build ~3000 others. My next cohort starts next week, on 28th March (coming weekend), and there are still ~7 seats left in case you are interested. Curriculum and other details - arpitbhayani.me/course

no greater happiness than getting your ideas patented in AI. Onwards and upwards.

One thing I love about the Anthropic team is how they ship a blog or tutorial with every feature launch. I wish more startups did this. For me it is still the #1 way to build trust with developers. So I built a workflow to make it easier for anyone. 5 Claude Code skills for DevRel content: /technical-tutorial for code-first guides /api-docs-audit to catch doc gaps before devs do /code-examples for multi-lang samples ready to copy /dev-newsletter to changelogs devs actually read /tech-social to launch threads without the fluff Adapted from @coreyhaines31’s marketingskills repo. Full API launch: 12–18 hrs → a few hours. Happy to get feedback and what more should I add here! github.com/techtrailhead/…

@morgan_twts School prayer. Our Father, in heaven Holy be your name..

what is something you memorised years ago and can still recite for no reason?

tried the Thai Coconut coffee for the first time. Worth a shot.

tried doing this with remotion a few days back. 1. Gave access to my own YT video. 2. Specified the timestamps. 3. Asked to cut it in vertical format. 4. Add BGM, thumbnail, subtitles, and other animations. Voila! A reel was created. Working with it for the first time took some extra efforts and hours. Once done, created a reusable skills file. The hassle of creating shorts/reels just dropped down.

@vps20571 Yes. Built already.

@ankitatr_ are u building?

your idea of opening an AI services agency just got easier. -/ pick a niche with anything <10% -/ build an agent with Claude to scrape you clients for that domain Skills + MCP -/ Extract all info -/ Build a DB -/ Automate the workflow for daily use and start pitching. (Created this entire workflow for myself; incredibly good)

@DrDatta_AIIMS Yes it is. I tried some use cases too and didn't work well.

@ankitatr_ I tried on my use cases

Tried chatjimmy and was amazed by the speed of response. I typed a question, and bam!answers flew out at ~14k tokens/sec on Taalas' HC1 chip. Although the speed is amazing the responses were inaccurate. I checked their blogs and a few other sources, to understand it's not the Llama 3.1 8B model failing but their v1's aggressive mixed 3-6 bit quantization to hardwire the whole thing into silicon for that insane speed. They prioritized proving model = hardware works end-to-end first (huge win on speed & efficiency). Still super impressive milestone. Can't wait for HC2 with better quant!

realisation is good. The next steps matter now. As a marketer, try leveraging AI now. > AI builds campaigns but you can experiment faster. Your edge now is the speed of iteration. > Build agents to scrape Google & Meta Ads Library. AI won't go get that for you unprompted. > Set up MCPs to auto-generate performance reports. Stop pulling numbers manually but start making calls instead. > Own your dashboards. Don't pick a dashboard tool. Build your own reporting app. Claude + Cursor + your ad API. (but yes, there are tools like Whatagraph, Databox to leverage) > The job exists only the expectations have shifted. > Ship creatives with what's available using Midjourney, Flux, Kling, Runway. I tried creating some sample image ads using ChatGPT image generation, my CTR went to Rs.7. Crazy metrics.

Link: open.substack.com/pub/bigthinkme…

have been digging deeper into the domain only to find this amazing video where Chip War author Chris Miller explains - how microchips are made, - why their production is so insanely hard to scale, and - why the world’s economic future may hinge on a technology most of us will never see.

@ninadwrites 🤣 It is very difficult to please them.

cooked veg dinner for a pure non-vegetarian friend and she was happy with the results. Verdict: Passed.