Uday Bondi

I'll start documenting my journey of showing up. I build tech I train MMA I lift Living with discipline. Sharing the process. Let's go 🚀

Log: #4 Deep work: 6 hours PM workout: S&C

Let’s get the reps in! Log: #3 AM workout: Jiu-jitsu Deep work: 4 hours PM workout: Active recovery and mobility

I'll start documenting my journey of showing up. I build tech I train MMA I lift Living with discipline. Sharing the process. Let's go 🚀

@giffmana @yacineMTB @iScienceLuvr I'm excited to try the checkpoint extracted from gemma3 and finetune it on custom data. Could you please point to best practices/code for finetuning siglip?

@yacineMTB @iScienceLuvr Dying to hear about it eventually! Also try the checkpoint extracted from gemma3 if you don't mind the resolution or finetune it.

Nice! They take the SigLIP encoder out of Gemma3 and further fine tune it on large amount of medical images to get MedSigLIP. This should be a solid image encoder for medical purposes anyone dealing with such images should give a try! (cc @iScienceLuvr) Why take SigLIP out of Gemma3? Because that is not plain SigLIP, but one that I specifically tuned further for Gemma3 and includes CapPa/LocCa/PaliGemma goodies, i believe it should be one of the best encoders out there. The "release downsampled model but is same weights" part is a little confusing though. Any of the authors care to expand on that?

@workflowwonder DELEGATE

@mervenoyann @giffmana A successor to the long reigning QwenVL! Very exciting. Im curious how the Qwen2.5VL compares to Apollo (7B variants)

@giffmana I'm a bit sad Qwen2.5VL slipped through the cracks amidst this release it's huge

Just had a quick look at DeepSeek's new Janus Pro paper. I don't think it's a big deal (yet...!), but quick TL;DR below before hype gets out of hands.

My go to OS vLLMs for videos right now: - Apollo - Qwen 2VL (will test 2.5VL) - CogVLM

Microsoft released LLM2CLIP: a CLIP model with longer context window for complex text inputs 🤯 TLDR; they replaced CLIP's text encoder with various LLMs fine-tuned on captioning, better top-k accuracy on retrieval 🔥 All models with Apache 2.0 license on @huggingface 😍

@sharifshameem That’s a great point. How would you go about objectively measuring them?

Video-Search is exciting. The generalizability of these models is mind blowing

@FarzaTV @_buildspace 🫡 Thank you for everything!

I have some tough news. Today, @_buildspace is closing. I wanna be real about everything. It’s not because of investor pressure, runway, internal drama, monetization, or any of those reasons. The actual reason is a bit more personal. I wrote a letter about it below. I know this company means so much to so many people. I hope after reading the letter, you can understand. Ty for the five best years of my life. No matter what happens, ily. buildspace.so/letter

@ramsri_goutham Congratulations Ramsri!

I'm excited to announce that I have joined the Knownwell .com team as a Lead Data Scientist, full-time! As many of you have been following my journey for years, I will answer all your questions with full transparency below :) But first, what impressed me about the Knownwell team? - They matched the salary I asked for with no negotiations. - The interview process was just three calls in total, focusing on alignment and fit. Two were with CXOs and one with the VP of Engineering. - There were no technical coding interviews, no statistics and probability questions, no take-home tests, and no problem-solving questions. - They were totally okay with me being remote in Hyderabad. Most well-paid jobs required me to be hybrid from Bangalore. - They didn’t have a problem with my involvement in my two other startups and didn’t view them as obstacles to my full-time job with them. - The work is really good. It is a perfect LLM + RAG + Agents technical problem that fits my interests! Now, coming to the questions that most of you might have: 1. What's your salary? I was looking for remote roles that fall in the range of $100k-$120k USD per year. I'm glad that this role meets that range! 2. What happens with your two MicroSaaS companies, Questgen. ai and Supermeme. ai? I continue to hold the same relationship with them as previously. They are stable and require very little maintenance from my end, so I don't view them as blockers of my time in any way. 3. Why take a job now? I have been in the AI SaaS space for the last five years, scaling two apps to ~750k+ users combined and $100k ARR. I have no desire to raise funds or take the aggressive high growth/marketing route. And in general, I feel the need for a change of pace and style given how long I have been in it! The blended 'Jobpreneur' role, given that the employer agrees to it, is the right transition. My child has also grown up to 3.5 years old, so there's more room for me to work for extended periods of time without much interruption, or to go to a coworking space and work. 4. Ramsri, will you share your hot takes as usual or become more diplomatic? The information that I shared above, like my salary, should give you a glimpse of how I operate internally. I will be mindful of any comments that might sabotage my company's reputation with other companies. But beyond that, expect me to be transparent and less diplomatic as usual. 5. Where did you find this role? Weekday .works is the only good portal I've found that lists well-paid job roles along with their salary ranges especially for remote Indian roles. If you have any more questions, ask away in the comments :) I will try to answer as transparently as I can!

@aribk24 Yes 🙌

going to be launching a new saas company in 2 weeks should I make a series on youtube on how I take the tool from 0 -> 100k mrr my goal is 100k mrr in <100 days

@louispereira Looks beautiful. I'm planning to work out of Goa for a while. I enjoy being close to nature, outdoor activities and meeting interesting folks! Would appreciate area recommendations in Goa :D

The monsoons are here and life is pretty good

@burkaygur Sounds exciting. I’m down to help ✋

We are assembling a team to speedrun the SD3 open source ecosystem in anticipation of next week's release: - SD3 finetuning scripts - SD3 Dreambooth - SD3 Adapters - SD3 Controlnets Who is in✋?

@s1ddok Satisfying indeed! What supplements are you taking?

there is something deeply satisfying about preparing your supplements for the week ahead

@taishik_ Fine tuning an SDXL model 👨‍🍳

What are you building this week, Japan?

@taishik_ @saviomartin7 Congratulations Taishi!!

Supabox has been acquired! It was generating some (very small) revenue, and I wasn't excited about solving email-related problems, so I was considering shutting it down. Then, something happened. I got a DM from @saviomartin7 He was very interested in the problem Supabox addresses and wanted to solve it himself. We exchanged a ton of DMs that late night. We agreed he'd take over Supabox, and we'd split the revenue. I knew his work from Twitter and PH. I actually spotted him at n&w s3 irl in Dubai (Savio, you were there right!?). He's an amazing builder and founder. I trust him to take Supabox further than I could. --- This isn't a life-changing success, but a series of small wins like this will build momentum imo. I'm truly grateful for the opportunity Savio has brought. Let's be positive and optimistic. Future is bright. LFG