Avi Zagorie

I am excited to share the latest blog, co-authored by myself, Junhwi Kim and Vyas Sekar, detailing how we managed to successfully localized and improved LLM by combining Fine-Tuning and RAG. https://lnkd.in/gKDkgGPq

70

3 Comments

Since launching last Fall, its been incredible to see all the many workflows and integrations that companies are building with Lucid's Developer Platform. We've seen companies build automated processes and deeply linked templates that help with Sales, Marketing, Operations, and Engineering teams get the most out of Lucid. Come learn how to get started building from two of our engineers on the API teams who help engineers build these integrations every day.

20

1 Comment

I've spent multiple semesters in the SAC and learned a lot about my own investment thesis, learning the mechanics of investing and meeting fascinating startups! If you're on the fence about angel investing, this is a good starting point

11

AI not as chatbot but making AI useful for customers.

7

1 Comment

🍜Anybody up for some ramen? I am excited to share that I am back in the garage, building again (it’s actually a house in the Berkeley Hills). We are taking the wraps off of Supercog AI, a new startup focused on solving Application Integration using LLM powered agents. My co-founder in this new venture is a great friend of mine (and a former co-founder partner), Alex Osborne. Alex and I have worked around the application integration space for a long time, and we are unrealistically excited about applying the GenAI stack to this problem. Not many people relish the task to get some bits to move from system A to system B. But it’s a critical job to be done, and that’s led to a lot of brittle scripts, a myriad of inflexible “no code” tools, and a lot of hand crafted SQL. But the power of the Large Language Model, trained on a huge corpus of information, offers a remarkable tool to solve this problem. The LLM you use today already knows the APIs of hundreds of popular systems. It knows the SQL dialects of every major database. It understands the semantics and data schemas of many popular SaaS systems. It understands much of the specific domain in which your business operates. I like to call the AI revolution “the last platform shift”. That may turn out to be hyperbole, but I definitely believe we are in the very early stages of seeing what this new stack can enable. I know we are hardly the first folks to claim that “amazing stuff is coming!”. But this is why Alex and I have decided to focus on a real and hard problem. If we can prove that LLMs can power a new way to solve this problem, it will be the first successful *new* approach in 15 years.. We will have much more to share about this new platform soon. In the meantime, if you’re interested in getting a peek - or even better if you have application or data integration tasks that you would like some help with - please reach out because we want to help. #genai #startups #backinthegarage

163

37 Comments

Last Tuesday, Malsgufa, our legacy CMS backend, was shut down. This milestone for BI's Tech and Product team culminated years of planning, effort, and execution. Malsgufa's first commit was on July 25, 2014, leading to 5,186 commits across nearly 30 repositories. Built and maintained by scores of engineers, its deprecation represents significant cost savings (20 production replicas, 3 instances of Solr, and an insane number of Zookeeper instances!), reduced security risk, and the elimination of an often-blamed boogeyman. The journey began in January 2020. Our challenge: moving traffic off Malsgufa without rebuilding everything in Muse. Initially, we tried GraphQL as an intermediary between Viking and our services, but progress was limited by Viking's complex Angular/Vue architecture. In January 2023, we pivoted to the strangler fig pattern with a shim. This approach proved far more effective – all Malsgufa traffic went through the shim first, with verbose logging to track unhandled requests. Meanwhile, we built Story Versioning API (SVAPI), incorporating lessons learned from Malsgufa's limitations around history and publishing states. We ran four major migrations (stories, custom pages, slideshows, and videos), building support in Muse and shimmed compatibility for Viking. Traffic gradually decreased from a torrent to a trickle, leading to the gloriously uneventful shutdown. Next, we will deprecate Viking by Q2 2025, unifying our users in the Muse ecosystem to create the industry's most innovative CMS. Special thanks to the entire Story Creation team: Chase Gruber, Chris Speziale, Daniel Arita, Danny Feliz, Dasha Lary, Dawin Camillo, Eric Saam, Frederica Chen, John Laslo, Kelly Filush, Morgan Cohn, Sarah Pai, Tyler Greenfield! Thanks also to DaD, Engagement, Helpfulness, Cloud Engineering, Test Engineering, and Operational Excellence! Finally, we recognize Malsgufa itself. Developed as a unified CMS backend to support Business Insider's expansion into new markets, it served our global newsroom reliably for a decade, publishing millions of stories despite no major changes in recent years. While it became less sustainable as our business evolved, it deserves our respect. Thank you, Malsgufa. May you rust in peace.

41

6 Comments

I was a guest at groCTO 🚀 a while back, talking about how to handle your #first90days as an engineering leader, establishing alignment among your teams and deploying #DORAmetrics and #Sonarqube for engineering accountability. Check it out! #learninpublic #engineeringleadership #onboarding #devcareers #learningmindset #growthmindset

26

2 Comments

𝗟𝗲𝘀𝘀𝗼𝗻𝘀 𝗳𝗿𝗼𝗺 𝗦𝘁𝗲𝘃𝗲 𝗝𝗼𝗯𝘀: 𝗧𝗵𝗲 𝗣𝗼𝘄𝗲𝗿 𝗼𝗳 𝗗𝗲𝘁𝗮𝗶𝗹𝘀 𝗮𝗻𝗱 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗲𝗱 𝗘𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲𝘀  The 𝘚𝘵𝘦𝘷𝘦 𝘑𝘰𝘣𝘴 biography by Walter Isaacson highlights how transformative a dedication to detail can be in building iconic products. Jobs was relentless about the small things – the often-overlooked details he believed were essential to a product’s soul. Take, for instance, his insistence on a slot-loading CD/DVD drive for the iMac instead of the traditional tray. For Jobs, this wasn’t just an aesthetic choice but a statement of simplicity and elegance. He felt users would sense this refinement, even unconsciously. It’s a reminder that what might seem minor can ultimately shape a product’s perception and feel. This meticulous attention extended across everything Apple created. Jobs was known to care about typefaces, button layouts, and even the arrangement of circuit boards inside devices, setting a standard that every touchpoint should be crafted with care and intention. In software, he pushed for fluid animations and subtle sounds that felt intuitive, creating harmony between user and product. But his focus wasn’t only on the details – he believed in building fully integrated experiences. Apple’s software ecosystem is a perfect example. Consider how iMessage, FaceTime, and AirDrop work seamlessly across devices. This seamless integration wasn’t accidental; it was a result of careful planning where every piece was designed to complement the others, allowing users to switch effortlessly between devices. Another example is iCloud, which connects photos, documents, and apps across devices in real-time, giving users a sense of continuity and connection. For software and digital products, these principles are powerful. Imagine creating an experience that doesn’t just optimize each part but feels unified end-to-end. Integrating design and functionality seamlessly – from backend processes to front-end interfaces and customer support – can turn a functional product into a memorable one. And paying close attention to the details, like smooth transitions or well-paced loading speeds, often makes the difference between a good experience and a great one. Jobs’ approach reminds us that a product’s magic often lies in the fine details and the effort to build an experience that feels complete, connected, and crafted with intention. #ProductDesign #CraftingExperiences #LeadershipLessons

4

2 Comments

A very timely and Insightful read from #hbr !! Tech is changing at a break-neck speed, product lifecycles are becoming shorter, and along with that, new challenges are coming up on the people management front as well. Today’s leader needs to experiment, fail fast, adopt and adapt faster and still keep the business moving in the positive direction. Go through this informative article and let us know some of the best practices that have worked for you in balancing the “traditional” techniques of mgmt with “emerging” needs.

22

1 Comment

Recently, Eric Ignasik and Michael Sols from The Software House asked me about my first steps as a VP of Technology at Habito. It's been a fun 8 months in a supportive environment with fantastic people! 😉 This interview discusses a few improvements ‌I made with a big emphasis on why servant leadership is important to me. Hope you’ll learn from it, ➡️ https://hubs.la/Q02Bw-Bv0 #softwaredevelopment #cto #softwaredelivery

63

4 Comments

Influencing to get things done - is what we do consciously and subconsciously everyday- at work, in the family and with friends. A lovely article from #hbr breaking down this important aspect of our interactions with a few specific use-cases and practical takeaways. Remember - influencing is not the forte of leaders only - we all can influence to win in whatever we do ! Take a read and jot down your own thoughts (or experiences) in the comments section. #influencewithintent #gettingthingsdone

29

5 Comments

This is another huge step forward for Appvance IQ. Since we introduced our AI Coverage Map in 2020, it has become a key feature in how AI Automation Engineers and their Dev counterparts strategically deploy unlimited AI testing resources targeting the most business-critical and vital user flows. The new coverage maps in 5.2 bring even more insight into how enterprise applications are really working and where defects are lurking, where performance is lagging, and where risks are undetected. Let me know if you'd like to learn more.

8

BREAKING: Aptos sets a new record on 8/14/24 with 326 million user transactions in a day (see https://lnkd.in/gbnfMGmx). That's an average of ~3766 user txns per second over a day, with fees as low as ~$0.00005 per txn. The network continued normal operations during the day without issues. This achievement more than doubles the previous record set (also by Aptos) of 154.4 million on 5/25/24 (~1787 txns per second over that day). The previous records were 95.6 million (Aptos) on 5/24/24 and another chain with 65.8 million in 7/23. Aptos now holds the top three records for daily transactions and is the clear infrastructure leader in mainnet throughput and speed in production. Remarkably, these transactions are not simple or trivially parallelizable either. They access global state (shared counters) that serialize transactions on other networks and are only parallelizable on Aptos. Web3 infrastructure is ready for the large-scale and novel use cases in the world. Major innovation to continue rolling out in the coming months!

338

18 Comments

I'll be at #PlatformCon24 in a few weeks to moderate a panel discussion on Architecting the Future: Platform Strategies for API Adoption & Impact. I am eagerly looking forward to learning more about how API leaders from Wayfair, The New York Times and Cox Automotive have led digital transformation with Apollo GraphOS to yield tremendous impact for their users and their business! RSVP if you'd like to join in person! #ApolloFederation #Platformengineering #APIOps #GraphQL

24

Parenting children and prompting GenAI are more similar than you might think. In a recent conversation with customer and good friend, Alan Tsang, we were discussing some of the challenges of working with GenAI. I shared a recent struggle I had with blatantly incorrect LLM responses, where no amount of prompt re-wording would generate the expected results. I told Alan “No engineering exercise in my 20 years of experience has felt this much like dealing with an unruly child.” My kneejerk reaction after trying 73 different prompts, was to get angry and berate the LLM for not understanding my perfectly clear instructions. That got me nowhere. Then it occurred to me that I should ask the LLM why it responded incorrectly:  • How did you decide to respond the way that you did?  • What training data did you use for this conclusion? • How can I instruct you differently to get the desired result? I immediately learned what specific training and prompting led to the LLM’s incorrect decision, and I was able to sever those connections in my final prompt, leading to perfect results. When I explained this process to Alan, he completed the analogy: “Hey, that’s exactly how you’re supposed to parent a misbehaving child! Instead of getting angry, ask them questions that help you understand why they made that decision. Then address the underlying cause.” So, whether you’re parenting or prompting today, don’t get angry. Ask the right questions and get to the heart of the issue.

32

8 Comments

Leaders, ask not what your reports can do for you but what you can do for your reports. Maximize your impact by fostering environments where people can do their best work. Great post from John Cutler about shifting a reactive perspective and getting clarity on how to respond when things seem too complex. #Leadership #ProductDevelopment #PeopleFirst #Empathy

5