Arhaan Aggarwal

Arhaan AggarwalSep 11, 2026

Students: want to learn AI by building real technology? We’re recruiting for AlgoBound Technologies next technical externship cohort, and you don’t need to be an advanced engineer to get started. We’re looking for students who are curious, take initiative, and want hands on experience solving real problems. At AlgoBound Technologies, we build software, AI tools, workflow automation, and data systems for companies. Our research also explores quantitative trading infrastructure and AI applications in high frequency trading systems. This program is a way to get involved, build your skills, and explore where you want to go technically. Here’s what the experience looks like: → Learn the tools, then use them. Everyone starts with our AI literacy curriculum: LLM fundamentals, context windows, prompting, coding agents such as Codex and Claude, and agent workflows. → Build at your level, and grow from there. Projects are matched to your experience and interests, from websites, UX, React dashboards, and automation to backend systems, databases, machine learning, cybersecurity, and AI engineering. → Get guidance along the way. Set concrete development goals, work with a mentor or project lead, and receive structured checkpoints and technical feedback. The program also includes workshops, client presentation opportunities, career exploration, and a certificate and digital credential. Everyone begins with consulting projects. Strong performers may progress into more specialized engineering or research work. Our broader network includes founders, engineers, researchers, and senior technical professionals with experience at Palo Alto Networks, Protect AI, PwC, AIG, Alphabridge, Lucrative AI, Idma Laboratories Ltd/ An Idma Group Company, Punjab Agri Ventures Ltd, and Y Combinator and Berkeley SkyDeck backed startups who will be joining us for exclusive panels. The commitment: Approximately 10 hours per week minimum, with flexibility around academic schedules. Interested in the upcoming cohort? Reach out at [email protected] to learn more. Know a student who would be a great fit? Send this their way. To apply, fill out the attached form: https://lnkd.in/gkFZJw_N

Arhaan AggarwalSep 12, 2026

What if the next generation of AI data centers wasn’t built on land? Scaling AI isn’t just about better models. It’s also about the power, cooling, land, and infrastructure needed to run them. I’m super excited to share version of Project NEPTUNE, a prototype my team and I have been building to explore offshore AI infrastructure. Our ambition is to move some of that infrastructure to our oceans while minimizing its environmental footprint, not just move the same problems elsewhere. But we didn’t want to stop at a 3D render. We wanted something we could actually try to break. NEPTUNE is a browser based, design stage digital twin that connects the equipment you see to simulated power, cooling, networking, and compute capacity. You can trip a cooling pump or cut a power feed, watch the simulated effects spread through the system, and test whether standby equipment or an alternate power path helps it recover. Then compare designs across entire experiments and see how much compute capacity was lost, how long recovery took, and how the options compare on power requirements and estimated equipment costs. The part I’m most excited about? Adding more equipment doesn’t automatically make a design the winner. Sometimes extra redundancy helps. Sometimes the simpler design meets the requirements for less. And sometimes none of the options passes. You can replay what happened, inspect the equipment involved, and export the results, not just watch an animation. This is still version 1 of a prototype, not a validated facility. We’ve added independent numerical checks, but real world feasibility and environmental impact still need to be tested. NEPTUNE starts with a prototype. The ambition is to build offshore AI infrastructure, not just simulate it. Getting there means establishing the engineering, economics, and environmental case. That’s the work ahead, and I’m excited to connect with people who want to help shape it. If you’re focused on AI infrastructure, energy, or deep tech, I’d love to walk you through what we’ve built, the assumptions we’re testing, and what a path toward a real world pilot could look like. Send me a message. Let’s talk about taking NEPTUNE beyond the browser. Explore NEPTUNE: https://lnkd.in/gvMdgAE9

Arhaan AggarwalSep 15, 2026

Astra has changed how we work with AI. And if you’re still walking it through every small step, you might be missing its biggest upgrade. It’s been a minute since OpenAI dropped GPT-6 Astra, and I’ve been having my fun with it. Everyone has their hot takes(especially people that don't understand AI or tech). My biggest takeaway is that it’s changing how we actually work with AI. For a long time, the norm was to break a task into small pieces, ask AI to do one, review it, and hand it the next. The model did the individual steps. We were still responsible for moving the whole thing forward. The biggest difference I’ve noticed is how it handles long horizon tasks: working toward a goal over many steps, without you having to keep telling it what to do next. If you’re still giving it the same small tasks, I can see why it might not feel dramatically different from Sol or Fable. Those were already very good at the way we were using them. The difference becomes much more interesting when you give it more of the job. Give it a goal and a basic direction, and it starts working toward it. Give it that goal, plus your domain knowledge and the steps you know matter, and it has a much better starting point. Then add the constraints like which decisions it can make, what it shouldn’t change, when it should ask you, and what a finished result needs to look like. That’s where, on well defined tasks, it starts feeling less like using a chatbot and more like delegating to an experienced teammate. You still need to check the work. But checking the result is very different from manually pushing every step forward. For structured tasks in Work, you can get what would previously have been days of output in a few hours. And with the context it already has about how I work, I’m spending less time explaining how I want the result presented. What’s interesting is that my expertise hasn’t become less useful. It’s become useful in a different place: defining the job, setting the boundaries, and judging the result. I’m spending less time telling AI what to do next, and more time getting clear on what I need done. The models are getting better. But we won’t get the full benefit by keeping our workflows exactly the same. What’s a task you now hand over in one go that you used to walk AI through step by step?