Mustanseer Sakerwala

Reply if you can!

@kishanvats03 @kyrodb @JeffDean @arpit_bhayani The db sounds fantastic... The side by side demo would clear a lot of doubts.... Looking forward to it

Stop investing in vector database wrappers. The infrastructure required for long-running autonomous agents didn't exist, until today. Introducing @kyrodb - An Autonomous vector database where cache is a first-class part of the retrieval kernel, not just an external acceleration layer. Today's vector DBs fail in production because they treat memory like static lookup tables. In read-heavy scenarios, cold-cache spikes violently increase P99 latency and degrade UX. Long-running autonomous agents simply can't survive on this. So we built a new caching mechanism called Hybrid Semantic Cache (HSC). It natively learns from your access patterns to predict your next-likely queries. No more cold cache spikes. 12 μs P50 overall on MS-MARCO dataset, with 73.5% cache hit rate, which means 3 out of 4 queries never touch cold storage. On ANN track: → Performing SOTA across all datasets. → The only vector DB to break 99.5% recall on the hardest dataset GIST-960 (99.91% @ 95 QPS). 1M vectors, 960 dimensions. → Massive reduction in embedding recomputation and compute costs, basically TCO for an enterprise. But fast retrieval is just step one. We are moving toward creating the Cognitive Data OS. A perfect solution seamlessly combining ultra-fast retrieval, right context, and reliable memory. This is what long-running autonomous agents need, this is what production systems need, and we at @kyrodb are building the bedrock for it. We launched our closed pilot last week, and the response has been incredible: companies are already building on KyroDB for agentic apps, real-time RAG, and voice AI agents. Now we're opening the pilot to a handful more teams worldwide. Retweet and comment 'kyrodb' to get early access.

@arpit_bhayani Congratulations 🥳🎉🎊

Joined Razorpay as Principal Engineer II :) From being a long-time customer to now building parts of the system - it's a full circle. Fintech is a new territory for me - time to get under the hood of how money actually moves. New domain, same guarantees - availability, correctness, performance - just with real money on the line.

@arpit_bhayani Hence proved, LLMs too cannot "read between the lines"

Here's a super interesting paper that talks about how the position of your relevant document in the prompt determines the quality of the output! This week, I am reading "Lost in the Middle: How Language Models Use Long Contexts," and it challenges a core assumption about long-context models. The setup is clever - they test models like GPT-3.5-Turbo and Claude on multi-document QA by changing where the relevant information sits in the prompt. If these models truly use their full context window, position shouldn't matter much, right? Wrong. They found a U-shaped performance curve - models perform best when relevant info is at the very beginning or end of the context, but performance drops significantly when the answer is buried in the middle. For GPT-3.5-Turbo with 20 documents, mid-context performance actually fell below its closed-book baseline (56.1%). Even on a synthetic key-value retrieval task, some models struggle to find exact matches in the middle of their context. If you are building RAG or work with long-context prompts, this paper will make you rethink how much context you're actually feeding your models; give it a read. You can always find the papers on my papershelf, and other write-ups in the comments. I will cover some of these topics in future videos. Until then, you can watch detailed explainer videos I made around system design, databases, and concurrency, again, linked below.

Mentoring someone is one of the most satisfying ways to share what you have learned and see someone grow because of it, but doing it well is what makes the difference. - share best practices, insights - try to make them more efficient with their work Be open with sharing your real-world lessons; that's what truly helps your mentee. Show them how to avoid pitfalls you have already seen. Give them tips to tackle tasks faster and with fewer mistakes. And when they stumble, be there. - provide emotional support if things go wrong - share truths rather than comforting lies It is easy to sugarcoat failures, but real growth comes from honest feedback, delivered kindly. When they feel overwhelmed, remind them that setbacks are normal. Stand by them so they know they are not alone when things break. - define short-term goals and chase long-term visions People tend to be either short-term doers or long-term thinkers, but as a mentor, help them balance both. Define clear, small wins, but always tie them back to the bigger picture so their efforts have a lasting impact. And above all, listen. Sometimes, your mentee doesn't need advice, but just an ear. Let them vent. Often, they will talk themselves into their answers, and that's growth worth showing up for. Great mentoring builds people.

@jetblack_ninja @arpit_bhayani I don't think today's nugget is just for professionals. The same applies to students, professionals and just all who live. This just makes an incredibly good culture around you. I Felt it, and I resonate with it.

@arpit_bhayani why am i reading this im not even 1 yoe

@manpurwala @Nike Thank you 😁

@DaScientist53 @Nike Many congratulations on your new role.

I’m excited to share that I’ve joined @Nike as a Software Engineer II! Looking forward to this new chapter. #Nike #justdoit

Feeling pumped to start this journey and be part of such an incredible team. Here’s to new challenges and opportunities ahead! 🚀 #SwooshLife #LifeAtNike

Special shoutout to Salma K, Deepa Shetty, Sheshadri Hejamadi, and Sanjay A. for making the hiring process smooth and enjoyable. Really appreciate it!

@theonemishra Though these are just my thoughts. It's explained differently in the book

@theonemishra And most of the times, it's much more. Every part of this is hard.

Delegating things are hard.

@theonemishra Also, this is a quote in well explained in: how to enjoy your life and your job by Dale Carnegie. It's a good read. And recently experiencing this, I felt it needs to be told more.

@DaScientist53 why? curious about the potential framework solutions as well.

Do drop by! @acanubhav will be there to demonstrate both Whatsease and @Krosspost_ai!

Testing in progress. We will be releasing this project soon. @TeamShiksha @sunnykgupta

The urge to leave everything and become a blacksmith.