Rija Javed

Hiring and working with excellent people is what makes starting a company so fun

16

Leading Cross-Org Tech Reviews Josie has written a great post about driving a cross-org multi-team tech review practice. It's very focussed, as only senior principal-level engineers are usually tasked with those projects. I do think that org leaders as well as senior engineers will benefit from reading the article and adopting the best practices described there. It's also an excellent example of how to scale one person's efforts across multiple team, and thus applies to more than just tech surveys. Project planning, roadmaps, and tech explorations all have the common characteristic of standardized templates, parallel execution, and open questions. Read on here: https://lnkd.in/gX6bnXzq

19

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

The policies being driven forward by Thierry Breton around the EU AI Act have been utmost damaging for the European tech industry, and they need to be revised as soon as possible. We need to better balance driving European innovation through data collection and modeling with conflicting needs such as data protection and safety. We need to ensure we can turn these developments into businesses that can be the pillar of the next generation of companies that provide productivity growth in Europe. We have no other choice - either we do this here, or we will have to take over products developed in the US and in China, as it already happened with search engines, cell phones, and social networks. We wouldn't want to let this happen with AI applications in robotics or digital assistants. #euacc #euaiact #aiact

9

💻Service-as-a-Software (SaaS) is the next AI breakthrough Move over traditional Software-as-a-Service—the new SaaS is here, where AI itself *is* the service. A game changer. I recently caught up with Varsha Ramesh Walsh, co-founder at Offstream (YC S24), who schooled me on this. Their platform is revolutionizing carbon project compliance. By automating the tedious regulatory layers, they free up developers to focus on what they do best—building biochar, direct air capture, and more. What are some other climatetech areas ripe for this new AI-driven SaaS revolution? Let’s discuss! #AI #SaaS #ClimateTech #M1C #CarbonTech

46

5 Comments

To say that I’ve had a bad week would be an under statement. I try to lose myself in reading and whilst it’s not the best coping strategy it does yield some interesting results … like this piece which really resonates with me: "Bad leaders execute, good leaders improve, and great leaders compound. Compounding is not about improvisation but a coherent set of actions in response to strategic challenges ahead. [...] A strategy is a designed response to the challenges ahead." We often fall into the trap of viewing strategy as a cure-all, neglecting to validate if the problems we identify are truly the most impactful for the business. We can create meticulously crafted documents brimming with well-intentioned policies and actions, but these efforts often yield minimal organisational recognition or lasting impact due to a lack of prioritisation. https://lnkd.in/ek8Xjfkx

15

Google/Nest needs to seriously throw the DIY community a bone, specifically Home Assistant users. It’s important to remember that like many products, features are often derivative works of the ingenuity and ideas of power users. So it was disappointing to learn today that despite us being all-in on Google household hardware (ie, if there’s a G product in the space we opt for that - 20+ devices across cameras, thermostats, and speakers), that because of that amount of adoption the ability to integrate with Home Assistant is rendered useless. How does the Device Access Sandbox API have a 5/10 QPM rate limit PER PROJECT Anish Kattukaran? The docs explicitly say the Sandbox is intended for personal use.

8

1 Comment

How good are Trainium and Inference clusters for online serving of production traffic? Check this blog from Trishul Chilimbi on how Rufus responses are powered with high throughput and resiliency. #Trainium #Neuron #AWS Amazon Web Services (AWS)

6

Recently I found myself wondering why Alexa was never really a true conversational assistant 😑… With the rise of LLMs and multimodality Alexa has never really been useful. A research scientist 👩‍🔬 who worked on the Alexa team took a stab at why Alexa missed the mark in the LLM era of Machine Learning. 🌎 - https://lnkd.in/e8SuUHs6 ℹ️ Alexa was in over 100 million homes, but there were intense guardrails in place that would take weeks for teams to get data for analysis or experiments. ℹ️ The data was poorly annotated and surprisingly they did not have decent computing environments to work with the data in. Which must be a joke given Alexa is built on top of AWS. ℹ️ Many teams ended up working on similar projects but did not want to work towards a common goal, in order to push their own performance metrics or avoid getting their teams broken up. ℹ️ Experiments were forced to have product related outcomes too early in their lifecycle. Even though things in the GenAI space moves fast, particular problems still need time to be solved sustainably. Ultimately given Amazon’s size, the amount of top talent and deep resources it is not surprising that they dropped the ball with Alexa and missed the chance to truly bring value to their 100 million plus customers. 🤕

1

When I started to work as a DevOps engineer, I quickly learned that our mobile developers 🤩 cared about build speed and reliability above all else. ❓ What is the return on investment on spending 10 minutes trying out the Bitrise Cache? The formula should multiply the number of mobile developers, the average number of weekly builds triggered by developers, the average build length, and the estimation of possible speed gain. 🚀 Teams quickly scale to the point where investing in cache optimization makes sense. Check out the free 30-day trial.

7

A couple of recent improvements to Konquest for us to shout about. Our flexible yet accessible calculation engine is already our pride and joy, but we just keep making it better. Firstly, we recently added support for bi-monthly and semi-annual schemes, ideal for those kicker schemes where you want to close out the year strong and have consultants make up for a shaky start to the year. https://lnkd.in/emeMvYUP Then today, we have just shipped support for schemes which are calculated into any number of weeks from (and back from) a specified start date. Want to run three week schemes starting on the 2nd Wednesday of the year - now you can with Konquest 🦍 https://lnkd.in/eeC_KM9G We're currently hard at work on the next set of enhancements, this time looking at ways you can penalise blank months and missed thresholds in a variety of different ways. Feel your current commission structure is not getting the best out of your teams? There's never been a better time to speak to us. #commissionaccomplished

14

2 Comments

I cancelled our team's subscriptions to Github Copilot & ChatGPT after 10 days with Cursor 🌟 What Sets Cursor Apart Seamless Onboarding: The transition was remarkably smooth, reminiscent of Chrome's early days. Cursor's team clearly understands that muscle memory can be a significant barrier to adoption, and they've crafted an onboarding experience that makes the switch a delight. LLM Flexibility: The ability to switch between different Large Language Models, bring your own, or use the default Claude 3.5 Sonnet is a game-changer. This flexibility proves to be a massive advantage, freeing users from being tied to a single AI model. Multi-file Editing with Composer: This highly anticipated feature delivers 80% of the time, and when it works, it's phenomenal. It explains code changes, shows inline diffs, and allows for selective application of changes. My favorite one - https://lnkd.in/g4aYJbJQ. Contextual Awareness: The ability to @mention files, folders, and invoke web searches has transformed how I approach parallel feature development. It's like having a hyper-aware coding companion. Intelligent Auto-complete The tab auto-complete function is surprisingly powerful, working seamlessly across lines and files. It often feels like it's reading my mind, significantly speeding up coding sessions. Smart Inline Editing Cursor's understanding of the existing codebase shines in its inline code editing capabilities. It's prevented me from creating redundant utility functions by suggesting existing, similar functions—a huge time-saver. 💡 Pro Tip: Embrace smaller, more frequent code iterations. Commit changes often; you never know when you might need to start over on a change set. 🔧 Areas for Improvement Composer occasionally has UI glitches, taking over the entire window or hiding its title bar. Familiarizing yourself with keyboard shortcuts can help navigate these issues. Sometimes there's a disconnect between Composer's suggestions and what it displays. 🤔 Is Cursor Worth the Hype? For me and my team, it's a resounding yes. We're transitioning to Cursor as our primary development environment. I've noticed a significant reduction in my use of standalone AI consoles like Claude or ChatGPT, as Cursor integrates these capabilities seamlessly into the coding workflow. 🔮 Looking Ahead This shift to Cursor is part of our broader strategy to embrace AI-native development practices. I'm excited to share more insights in future posts about how we're leveraging AI to enhance our development processes. I'm curious to hear from other developers: What are your favorite Cursor features or hacks? How has it impacted your workflow? #DeveloperProductivity

22

4 Comments

It's becoming more and more clear that the future of development includes AI to co-write your code. Humans will be doing more Software Engineering/Development and less coding. Note that there are still significant limitations with having AI modify existing code without human help. Also, AI isn't great at systems-level thinking, cross-file/module thinking, edge-case thinking, etc. If you can define the functions you need, AI is REALLY good at implementing your functions today. Reminds me of some Masters work I did in 2000 where we were defining requirements using formal mathematics and then generating code based on the definition of the function. Only, it's MUCH easier to define behavior today in natural language. (Formal mathematics is hard)

8

1 Comment

It's official! The London Observability Engineering Meetup is back for an exciting talk about scaling and managing SLOs at a global scale! 🌏 Join us on Thursday, July 11th, where Alex, Senior Site Reliability Engineer at Google, will present the evolution of Service Level Objectives (SLOs) for the GCE Compute API over the past eight years. He'll start with the initial 30 SLOs, move through a phase with around a thousand, and end with millions of per-customer SLOs. He'll share anecdotes, techniques for handling low-QPS (continuous over discrete metrics), and strategies for aggregating data to enhance leadership visibility. He'll also give practical tips for running and improving this system in production. #observability #engineering #devops #sre

8

1 Comment

When innovation and leap-frogging competition is direly needed, the lack of technical prowess in mgmt layer is most acutely felt. Because, they often hold the most business context to match a business problem to a technical solution, having technical wherewithal is paramount. Don’t outsource the technical thinking to your team. Be that manager.

10

1 Comment

This has absolutely been my experience building prod LLM-based apps. It requires a change in mindset from ‘this passes all the tests’ to ‘this passes enough of the tests often enough and is valuable enough that we’ll work around the times it fails’. It also creates some weird dynamics in terms of SDLC because as Brian correctly points out creating an exciting prototype is child’s play and whereas for normal apps the prototype might represent 50% or more of the work required to ship with LLMs the prototype can do maybe 80% of what you want but the effort remaining to get it to 99% effectiveness is 100x what it took to create the POC. We’re all having to learn new ways of thinking about dev if we want to work productively in this space.

10

2 Comments