Siddharth Bakshi

N Prasad S. delivered a thought-provoking session to round out the day at Testing Talks Conference Sydney with his talk, “Quality Engineering in the Era of the Fourth Industrial Revolution.” Drawing on his experience at Tyro Payments, Afterpay, and Atlassian, Prasad explored the evolving role of Quality Engineers in an AI- and data-driven world. From ensuring data integrity for analytics and AI models to harnessing the power of DataOps and predictive quality, he outlined a bold roadmap for building resilient, future-ready systems. A compelling talk that challenged traditional thinking and inspired new approaches to quality in the digital age. #TestingTalksConference

42

2 Comments

Introducing srcery ✨ - a simple tool that I developed over the last few days in my spare time, in collaboration with Monika Rangasamy. https://lnkd.in/gWRcPf5k Sometime back, I had to analyze some legacy PHP code and extract all the features. As it was a perfect use case for an LLM, I tried a few tools that would flatten the source code, but I wasn't quite impressed. I wrote my own and was able to generate a detailed CSV of all the features! One of the reasons why this works well is the large context window (1M Tokens) of LLMs such as Gemini. One million tokens can represent approximately 30,000 to 75,000 lines of code. You can do a lot with that! Here are the current features of srcery: 🔸 File Consolidation: Merges multiple source code files into a single output file along with the source tree. 🔸 Prompt Creation: Can analyze a project's source code and create prompts to provide an overview, create a feature list, or generate a README (srcery's README was generated by itself :)) 🔸 Customizable: Supports custom file inclusions and exclusions through config.json to tailor the output to your needs. 🔸 GitHub Integration: Can clone and process a GitHub repository directly from a URL. 🔸 Project Type Detection: Automatically detects common project types (Node.js, Python, Java, .NET) and applies relevant default settings for file inclusion and exclusion. Try it out and let us know your feedback. If you want to participate and build new features, feel free to send a PR our way!

153

5 Comments

Debugging is where real development begins. Code may compile, but only debugging reveals the truth behind its behaviour. Debugging Mindset: Think Like a Bug Hunter 🔍 Key lessons we all learn the hard way: 1. “Works in DEV” doesn’t mean “Works in PRD” 2. Small data hides big design flaws 3. Uninitialized variables are silent time bombs 4. FOR ALL ENTRIES: hero in dev, villain in PRD if not handled properly 5. Update task debugging reveals what normal debugging hides 6. A short dump is not the problem - it’s only the messenger Real skill is not just writing code - it’s questioning assumptions, reading between lines, and testing beyond the happy path. Finding the bug is half the battle; understanding why it was there wins the war.

40

8 Comments

It was wonderful meeting you Radhakrishnan Selvaraj! Liked your ideas like "Zero maintenance engineers", "Zero maintenance team", etc( I'm a big of putting out names for concepts and patterns so that it can be effectively communicated and reused). And thanks a lot for the signed copy of your book! We'll explore on various aspects of the "Zero maintenance" ideas!

42

🚀 Exciting! Infosys and Telstra have joined forces in a game-changing joint venture to accelerate AI-driven cloud and digital transformation for Australian enterprises! 🌏 Infosys will acquire a 75% stake in Versent Group, a leading digital transformation provider and Telstra subsidiary, for ~$153M. This powerhouse collaboration combines Telstra’s connectivity, Versent’s local expertise, and Infosys’ global scale, leveraging Infosys Topaz (AI-first solutions) and Infosys Cobalt (cloud offerings) to deliver secure, resilient, and innovative solutions. 💻✨ With a team of 650+ engineers and a focus on industries like government, finance, energy, and utilities, this venture is set to redefine digital innovation in Australia and New Zealand. The deal is expected to close in the second half of FY 2026, pending regulatory approvals. Here’s to driving a connected, intelligent, and digital future! 🚀 #AI #Cloud #DigitalTransformation #Infosys #Telstra #Innovation

3

Congratulations Amazon Web Services (AWS) 🚀 Big News from AWS! 🌏 Amazon Web Services has officially launched its new Asia Pacific (New Zealand) Region in Auckland with three Availability Zones (AZs) & API name ap-southeast-6. This launch marks a major milestone — giving New Zealand organizations the ability to host their workloads locally, ensuring data residency, low latency, enhanced resilience, and sovereignty-by-design. ✨ Key Highlights: 💰 NZ$7.5 billion investment in New Zealand ⚡ 100% renewable energy powered by the Turitea South wind farm 🌱 🏢 Trusted already by leading organizations like Kiwibank, Deloitte, Xero, TVNZ, University of Auckland, and many more 🎓 Commitment to train 100,000 New Zealanders in cloud skills by 2027 (50,000+ already trained!) With this launch, AWS now operates 38 regions and 120 Availability Zones worldwide, strengthening its global cloud footprint. This is a game-changer for innovation in New Zealand — unlocking new opportunities across finance, energy, education, government, and tech startups. #AWS #CloudComputing #NewZealand #Innovation #Cloud #AWSRegion #DigitalTransformation #amazonawsservices

21

Nice weekend read on comprehensive analysis on the two paradigms of AI assisted software engineering - Vibe coding and Agent coding.. Refer - https://lnkd.in/gYSvR8Bh

7

AWS Asia Pacific (New Zealand) Region is now open The new AWS Asia Pacific (New Zealand) Region gives customers more choice to run workloads and securely store their content in New Zealand while serving end users with even lower latency. The new AWS Region will give developers, startups, entrepreneurs, and enterprises, as well as financial services, retail, education, government, and nonprofit organisations, greater choice for running their applications and serving end users from data centres located in New Zealand. https://lnkd.in/gUHk2KqA

4

Enjoyed collaborating with Josephine Joyce on this DZone article: https://lnkd.in/gAhNzXd9 . In Performance-Focused Platform Engineering, we explore why platform teams must evolve beyond enablement and start engineering for measurable performance outcomes. Curious to hear how others are thinking about this. #PlatformEngineering #DevOps #SRE #CloudNative #IDP

8

I didn’t realise I was losing credibility with my own engineering team until the Slack messages started drying up. Not the big escalations. The small ones. The quick architecture sense-checks. The “does this approach feel right?” questions. The early warnings senior engineers usually raise when something looks slightly off before it turns into a real problem. Those messages didn’t stop overnight. They faded gradually, and I was busy enough not to notice. By the time I did, the gap had already started to show. At the time, I was managing a 25-person engineering team and spending most of my week in meetings. I still had the title. I was still involved in technical decisions. But the reality was simple: I was no longer close enough to the code, the systems, or the day-to-day engineering trade-offs for people to feel I had the right context. So they stopped pulling me in. It wasn’t disrespect. It was practicality. From their point of view, I was busy, operating at a higher level, and further away from the detail. Why interrupt the CTO with something small? That’s exactly the problem. The small issues are where technical leadership creates the most value: a poor abstraction, unclear ownership, a requirement that doesn’t make sense, a design choice that feels wrong before anyone can fully explain why. These are cheap to fix early and expensive to fix later. When those signals stop reaching you, things often look calm on the surface. In reality, that’s usually when risk is building underneath. One thing I’ve learned is this: authority and technical credibility are not the same thing. You can keep the title long after you’ve stopped being useful in the moments that matter. The answer isn’t more meetings or better status reporting. It’s visible technical presence. Reviewing the risky PRs. Jumping into design discussions. Building a reference implementation. Making space for the half-formed conversations where engineers test ideas, raise concerns, and surface problems early. If you lead engineering teams, ask yourself this: When was the last time an engineer brought you a small problem without being asked? If you can’t remember, it’s probably worth paying attention. #EngineeringLeadership #TechnicalLeadership #SoftwareArchitecture

24

11 Comments

Here is an interesting chat with Claude4.0 today morning. Do read it if you have time - https://lnkd.in/giWaxUn8 I have been helping customers improve their business processes using #generativeAI for over 2 years but some of the latest advancements lead me to humbly think - if we are going in the right direction specifically around #superintelligence and its impact to #society and #environment. Conversation is interesting noting Claude's responses of how our brain works and its similarity to large neural networks, #consciousness and its #selfawareness. Even though I know it's just mathematical computations - and I disagree to call it "intelligent" - still intrigued by the way it's framing responses. Note the part where it went on to talk about GPT models and impact to environment but not its own :-) (in a very human way). To summarize - 1) organizations developing this technology should start thinking beyond immediate gains and into the future of humanity, not fall into the trap of being first when it comes to AI - but develop it carefully, responsibly and with considerations to the future. 2) To people (like me), I would urge us to stop using these models "just for fun" - specially generating videos or massive artifacts if it does not relate to your business/work. One video generation uses power equivalent to at least 10 hours of continuous microwave operation - which is just the consumption part - the power needs are compounded by underlying server cooling etc. Let's do our bit. The thoughts on this post and attached chat are of my own and not my employer.

9

1 Comment

System Design Part 2 Continuing my System Design Blog Series, this post is a deep dive into the Low-Level Design of WhatsApp. 📱 From handling 2 billion daily messages to ensuring high availability, we’ll walk step by step through how WhatsApp is architected at scale. 🚀 High-level estimate Here is our high-level estimate: TypeEstimateDaily active users: (DAU)50 million Requests per second (RPS) : 24K/s Storage (per day) : ~10.2 TB Storage (10 years) : ~38 PB Bandwidth : ~120 MB/s

6

𝗧𝗵𝗲 𝗦𝗼𝘂𝗻𝗱 𝗼𝗳 𝗦𝗶𝗹𝗲𝗻𝗰𝗲: Giving Voice to the Unspoken with AI 🤫🎙️ I recently watched the silent film 𝗚𝗮𝗻𝗱𝗵𝗶 𝗧𝗮𝗹𝗸𝘀 and was struck by the power of non-verbal communication. It sparked a question: What if we could give a voice to silent lip movements in real-time? So, I built a "𝗦𝗶𝗹𝗲𝗻𝘁 𝗟𝗶𝗽𝘀-𝘁𝗼-𝗦𝗽𝗲𝗲𝗰𝗵" engine to find out. 🎬 𝗧𝗵𝗲 𝗦𝘂𝗽𝗲𝗿𝗵𝘂𝗺𝗮𝗻 𝗠𝗼𝗺𝗲𝗻𝘁: To test it, my daughter and I sat in front of the webcam and simply... mouthed words. No sound. No recording. Just live, silent speech. The results were genuinely surprising. As we spoke silently, the program translated our lip movements into text and spoke them out loud. It felt like a real-life version of the movie 𝗦𝗵𝗮𝗺𝗶𝘁𝗮𝗯𝗵—where technology lends a majestic voice to a silent artist. Or imagine a cliffhanger in 𝗦𝘁𝗿𝗮𝗻𝗴𝗲𝗿 𝗧𝗵𝗶𝗻𝗴𝘀, where a character mouths a secret warning that only an AI can decode before the villain notices. 🚀 𝗕𝗲𝘆𝗼𝗻𝗱 𝗦𝗶𝗹𝗲𝗻𝗰𝗲: While silent speech is fascinating, the real-world application is even bigger. Imagine combining this visual data with standard audio models like 𝗪𝗵𝗶𝘀𝗽𝗲𝗿. In a noisy cafe or a office workspace, audio-only models struggle. But by fusing lip movement data with audio (Audio-Visual Speech Recognition), we can achieve incredible accuracy even in high-noise environments, all running on local hardware. It’s like giving AI a pair of eyes to read what it can’t clearly hear. ⚙️ 𝗧𝗵𝗲 𝗧𝗲𝗰𝗵 𝗦𝘁𝗮𝗰𝗸: I wanted this to be a fast, local and cost efficient experiment, so I avoided heavy cloud APIs. Here is the pipeline: 𝗧𝗵𝗲 𝗘𝘆𝗲𝘀 (𝗩𝗶𝘀𝘂𝗮𝗹 𝗦𝗽𝗲𝗲𝗰𝗵 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻): I used 𝗖𝗵𝗮𝗽𝗹𝗶𝗻, which leverages MediaPipe to track facial landmarks. It isolates the lip region with incredible precision to predict phonemes from the live video feed. 𝗧𝗵𝗲 𝗕𝗿𝗮𝗶𝗻 (𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗖𝗼𝗿𝗿𝗲𝗰𝘁𝗶𝗼𝗻): Raw lip-reading is noisy ("Water" looks like "Wall"). I piped the raw text into a local Small Language Model (SLM) running on Ollama. Crucially, I injected a custom dictionary into the prompt context. This allowed the model to bias its predictions towards specific vocabulary, resolving visually identical words (𝗵𝗼𝗺𝗼𝗽𝗵𝗲𝗻𝗲𝘀) with surprising accuracy. 𝗧𝗵𝗲 𝗩𝗼𝗶𝗰𝗲 (𝗟𝗼𝗰𝗮𝗹 𝗧𝗧𝗦): Finally, the purified text is spoken aloud using local voice engine using 𝘀𝗮𝘆 commands in MacOS. I’ve also been experimenting with other TTS like Kokoro, Piper TTS and advanced premium voices in MacOS. While ElevenLabs remains the gold standard for realism, these local models are becoming shockingly good at zero latency. But that's a talk for another day! Code: https://lnkd.in/g45Ke-g8 Has anyone else been experimenting with AVSR or local TTS models? I’d love to hear your thoughts! 👇 #GenAI #LocalLLM #ComputerVision #VSR #SpeechRecognition #AI #DeveloperProductivity #TechInnovation #GandhiTalks #Shamitabh #NeuralInterfaces

58

3 Comments

Crocnex Limited will track several key performance indicators (KPIs) to ensure progress towards its long-term vision, including revenue growth, customer acquisition rates, user engagement metrics, product innovation timelines, market share growth, customer satisfaction scores, and sustainability impact assessments. These KPIs will provide a comprehensive view of the company's performance and help align its strategic initiatives with its goals in the tech industry.

4