C
Cubiczan
Guest
Oh hey. I'm Shyam Desigan (most people call me Sam). I spent twenty-plus years in finance, mostly as a CFO. Now I'm a Director of Technology Consulting, and I learn the tech stack by building real things with AI tools, usually against a hackathon deadline.
I'm Director, Technology Consulting at Tiger Analytics in New York. I moved into the role in September 2026 after a long career in finance leadership. Outside the day job, I build. Since February 2026 I've submitted more than 35 projects to hackathons on Devpost, plus a steady stream on DoraHacks, lablab.ai and others. My favorite part is that these two halves feed each other. A business problem I understand from years of running finance becomes something I can actually prototype, and every build teaches me a tool or a layer of the stack I'd otherwise only read about.
Honestly, I came in through the finance door. After my MBA at Indiana University's Kelley School of Business, I spent my career in finance. My first official tech title came at the American Academy of Physician Associates, where I was CFO and Senior VP of IT, so the systems and the books were both my job. After that came KIC Ventures, Sproxil, TechCXO, Purim Ventures, BlackTeal Energy and Green Li-ion, and every role pulled me further into data, automation and systems. What changed recently is that AI tools made it realistic for someone like me to build software, not just spec it. My first hackathon submission I can find is a Commodity Price Analyzer for the Airia AI Agents Hackathon in February 2026. Within a few months I had won a prize at DeveloperWeek New York and an award at Mantle's Turing Test Hackathon, and that convinced me the approach worked.
Green, but the green of a runway dashboard, not a flag. In my world green means the cash lasts past the next milestone, the risks are known and nobody is panicking. A utopia would be a place where most people's "status light" is green: enough runway to try things, fail and try again. That's also the setup I'm trying to give myself as a learner.
I'll take the second one. My favorite thing is reading builder write-ups by people who explain what broke and how they fixed it, not just what they launched. That kind of honest postmortem has taught me more than most polished tutorials. As a newcomer to building, it helps to see that experienced engineers also hit dead ends, file bug reports and work around flaky tooling. HackerNoon has also been part of my building directly. My project CHP Locknet was shortlisted for the Decentralized Compute Track of the HackerNoon × Nosana Decentralized AI Hackathon. I'd like this interview to be my first contribution in the write-up spirit.
I treat hackathons as a forced curriculum. Each one drops me into a different tool and a different layer of the stack, with a deadline that keeps me honest. These are the builds I'm proudest of:
I don't do this alone. I work with a small team of AI agents (assistant bots) that handle research and parts of the builds, and our code lives under the GitHub org icohangar-ops. I act as the CFO of that team as much as its developer. I decide what's worth building, check the output and own the result. I also run on hackathon and startup credits instead of cash, so I'm spending time, not budget, while I learn.
Anyone can show up and contribute. A finance guy can open a pull request on an open-source project, and if the code is right and CI is green, it gets reviewed on the merits. Hackathons work the same way. Nobody asks for your résumé, they look at your demo. That openness is the whole reason my learning plan works.
A solar-charged laptop loaded with offline models and documentation. A laptop is a workshop, a library and a ledger in one box, and with a local model it's also a patient tutor that doesn't need the cloud. Someone would still need to track the group's supplies, and old habits die hard, so I'd probably build the zombie-era runway dashboard on day one.
The noise, especially in finance and crypto, where confident-sounding promises are everywhere and proof is rare. I've learned to trust things I can reproduce: a running app, a passing test, a bug report someone else can follow step by step. That's partly why the ChainSight AI award meant so much to me. It was for a deployment that actually runs, not a slide. The internet rewards loud claims too often. I'd rather be the person who ships the small working thing.
I'd put it into the unglamorous plumbing of critical minerals: recycling, traceability and the financial infrastructure that helps smaller mining companies get to production. I worked on battery recycling at Green Li-ion, building data and systems for commodity pricing and traceability. I've also built hackathon tools in this space, including CritMin Compass and Mineral Watch Agent. Again and again, good projects stall because of capital access and poor data, not geology. I'd pair the capital with agentic tools for diligence and reporting, so smaller companies can look as investable as they really are.
Right now it's agent engineering for real: guardrails, evaluation and running autonomous systems that touch real money and real data. ChainSight AI taught me how much work goes into trusting an alert. Council taught me that several agents checking each other beats one confident agent. Once agents can act, the interesting question is what stops them from acting badly. I'm also learning the parts of the stack I used to hand off, like infrastructure, CI, GPU compute and on-chain plumbing. Each hackathon adds one more layer, and that steady pace matters more to me than any single project.
The future. Ten years ago I would have needed a full engineering team to build what I can now prototype in an afternoon, and I don't want to give that up. I'm curious whether finance work, which still runs on email, data rooms and spreadsheets, will look completely different by then. Part of me wants to go check whether the things I'm building now turned out to matter.
Practical and grateful, with a CFO's skepticism built in. AI didn't make me an engineer overnight. It gave me a fast enough feedback loop that I could become a builder by shipping, breaking things and fixing them in public. Most of my hackathon entries didn't win, and that's fine, because the two that did came out of everything the others taught me. The catch is that AI output is a first draft, not a finished product. Someone still has to read the code, test the edge cases and take responsibility for the result, just like signing off on a set of financials. My advice to other non-engineers is simple: pick a real problem you understand, enter a hackathon with a deadline and let the tools teach you the stack one layer at a time.
Connect with me on LinkedIn (linkedin.com/in/sam-desigan-909454327) or GitHub (github.com/icohangar-ops).
1. What do you currently do and what's your favorite part about it?
I'm Director, Technology Consulting at Tiger Analytics in New York. I moved into the role in September 2026 after a long career in finance leadership. Outside the day job, I build. Since February 2026 I've submitted more than 35 projects to hackathons on Devpost, plus a steady stream on DoraHacks, lablab.ai and others. My favorite part is that these two halves feed each other. A business problem I understand from years of running finance becomes something I can actually prototype, and every build teaches me a tool or a layer of the stack I'd otherwise only read about.
2. How did you get started with your Tech Career?
Honestly, I came in through the finance door. After my MBA at Indiana University's Kelley School of Business, I spent my career in finance. My first official tech title came at the American Academy of Physician Associates, where I was CFO and Senior VP of IT, so the systems and the books were both my job. After that came KIC Ventures, Sproxil, TechCXO, Purim Ventures, BlackTeal Energy and Green Li-ion, and every role pulled me further into data, automation and systems. What changed recently is that AI tools made it realistic for someone like me to build software, not just spec it. My first hackathon submission I can find is a Commodity Price Analyzer for the Airia AI Agents Hackathon in February 2026. Within a few months I had won a prize at DeveloperWeek New York and an award at Mantle's Turing Test Hackathon, and that convinced me the approach worked.
3. If Utopia were a color what color do you think it'd be and why?
Green, but the green of a runway dashboard, not a flag. In my world green means the cash lasts past the next milestone, the risks are known and nobody is panicking. A utopia would be a place where most people's "status light" is green: enough runway to try things, fail and try again. That's also the setup I'm trying to give myself as a learner.
4. If everything about HackerNoon changed drastically, what is one detail you'd like to keep exactly the same? OR What's your favorite thing to do with HackerNoon and why?
I'll take the second one. My favorite thing is reading builder write-ups by people who explain what broke and how they fixed it, not just what they launched. That kind of honest postmortem has taught me more than most polished tutorials. As a newcomer to building, it helps to see that experienced engineers also hit dead ends, file bug reports and work around flaky tooling. HackerNoon has also been part of my building directly. My project CHP Locknet was shortlisted for the Decentralized Compute Track of the HackerNoon × Nosana Decentralized AI Hackathon. I'd like this interview to be my first contribution in the write-up spirit.
5. Tell us more about the things you write/make/manage/build!
I treat hackathons as a forced curriculum. Each one drops me into a different tool and a different layer of the stack, with a deadline that keeps me honest. These are the builds I'm proudest of:
- ChainSight AI: Mantle's Turing Test Hackathon 2026 (on DoraHacks). A real-time monitoring layer over Mantle's DEX ecosystem. It flags anomalous on-chain behavior, such as flash-loan and sandwich attacks, wash trading and whale accumulation. Then an LLM turns each flag into a plain-English risk note. It won the Project Deployment Award (1,000 USDT) as one of Mantle's 30 winning teams. That award was for shipping something running in production, not just a demo, which is what I'm proudest of. It was also my crash course in on-chain data, Next.js, TypeScript and Prisma.
- Council Tower Pipeline: DeveloperWeek New York 2026 Hackathon. A multi-agent AI deliberation engine built on Tower's data pipeline. It won the "Tower Pipeline Challenge: Data-to-AI" prize. An earlier version, Council, was a multi-agent deliberation platform I'd built for the Vercel v0 + AWS Databases hackathon. Reusing and hardening it is how I learned that the second build is where the real learning happens.
- CHP Locknet: HackerNoon × Nosana Decentralized AI Hackathon. It was shortlisted for the Decentralized Compute Track and came with GPU compute credits on Nosana's decentralized network. That was my first time thinking seriously about where inference actually runs and what it costs per hour.
- P2P Copilot: UiPath AgentHack 2026. An AI-powered procure-to-pay pipeline, which is a CFO's home turf in automation form. It went through judging but didn't make the 31-team finale out of the 200-plus projects judged. Honestly, that was useful too. The tools I learned there helped me solve a real Outlook access problem at work afterward.
- The enterprise-cloud circuit. I built Working Capital Optimizer for the Google Cloud Rapid Agent Hackathon, MeshCFO Control Tower for the Nebius × NVIDIA Global AI Hackathon, and Settle, an agentic patient-billing assistant over RCS, WhatsApp and SMS, for the AWS CDS Agentic AI Partner Hackathon. No prizes there that I know of, but each one forced me into a different cloud, model provider and messaging stack.
- The long tail. There were also entries for Splunk, CockroachDB × AWS, DataHub, GitLab, Slack, Arm, Backblaze, Auth0, PayPal, OpenCV and RevenueCat's Shipaton, plus lablab.ai events with AMD, IBM Bob, AssemblyAI and Alpaca. Most didn't place. Shipaton told me plainly that my app wasn't selected. Each of them still added one more tool to the kit.
- Lately: this month I've been doing quicker, smaller challenges. RunwayPilot is a no-code fractional-CFO runway dashboard I built for the Base44 Challenge. I also built testnet trading bots for Somnia's dreamDEX bounty. Next up are the Open Agent Hackathon and the Meta Global AI Developer Hackathon, where my application was approved.
I don't do this alone. I work with a small team of AI agents (assistant bots) that handle research and parts of the builds, and our code lives under the GitHub org icohangar-ops. I act as the CFO of that team as much as its developer. I decide what's worth building, check the output and own the result. I also run on hackathon and startup credits instead of cash, so I'm spending time, not budget, while I learn.
6. What's your favorite thing about the internet?
Anyone can show up and contribute. A finance guy can open a pull request on an open-source project, and if the code is right and CI is green, it gets reviewed on the merits. Hackathons work the same way. Nobody asks for your résumé, they look at your demo. That openness is the whole reason my learning plan works.
7. It's an apocalypse of 'walking dead' proportions and you can only own a singular piece of technology, what would it be?
A solar-charged laptop loaded with offline models and documentation. A laptop is a workshop, a library and a ledger in one box, and with a local model it's also a patient tutor that doesn't need the cloud. Someone would still need to track the group's supplies, and old habits die hard, so I'd probably build the zombie-era runway dashboard on day one.
8. What is your least favorite thing about the internet?
The noise, especially in finance and crypto, where confident-sounding promises are everywhere and proof is rare. I've learned to trust things I can reproduce: a running app, a passing test, a bug report someone else can follow step by step. That's partly why the ChainSight AI award meant so much to me. It was for a deployment that actually runs, not a slide. The internet rewards loud claims too often. I'd rather be the person who ships the small working thing.
9. If you were given $10 million to invest in something today what would you invest in and why?
I'd put it into the unglamorous plumbing of critical minerals: recycling, traceability and the financial infrastructure that helps smaller mining companies get to production. I worked on battery recycling at Green Li-ion, building data and systems for commodity pricing and traceability. I've also built hackathon tools in this space, including CritMin Compass and Mineral Watch Agent. Again and again, good projects stall because of capital access and poor data, not geology. I'd pair the capital with agentic tools for diligence and reporting, so smaller companies can look as investable as they really are.
10. What's something you're currently learning or excited to learn?
Right now it's agent engineering for real: guardrails, evaluation and running autonomous systems that touch real money and real data. ChainSight AI taught me how much work goes into trusting an alert. Council taught me that several agents checking each other beats one confident agent. Once agents can act, the interesting question is what stops them from acting badly. I'm also learning the parts of the stack I used to hand off, like infrastructure, CI, GPU compute and on-chain plumbing. Each hackathon adds one more layer, and that steady pace matters more to me than any single project.
11. Would you rather travel 10 years into the past or 10 years into the future? Give reasons for your answer.
The future. Ten years ago I would have needed a full engineering team to build what I can now prototype in an afternoon, and I don't want to give that up. I'm curious whether finance work, which still runs on email, data rooms and spreadsheets, will look completely different by then. Part of me wants to go check whether the things I'm building now turned out to matter.
12. How do you feel about AI?
Practical and grateful, with a CFO's skepticism built in. AI didn't make me an engineer overnight. It gave me a fast enough feedback loop that I could become a builder by shipping, breaking things and fixing them in public. Most of my hackathon entries didn't win, and that's fine, because the two that did came out of everything the others taught me. The catch is that AI output is a first draft, not a finished product. Someone still has to read the code, test the edge cases and take responsibility for the result, just like signing off on a set of financials. My advice to other non-engineers is simple: pick a real problem you understand, enter a hackathon with a deadline and let the tools teach you the stack one layer at a time.
Connect with me on LinkedIn (linkedin.com/in/sam-desigan-909454327) or GitHub (github.com/icohangar-ops).