Venturing with Vishesh — Startup Founder Interviews on SaaS, AI, Fundraising & Venture Capital.

# Hong Kong vs Shenzhen vs Singapore for Hardware Founders## Guest IntroductionI'm joined by Charudatta Datar, CEO & Co-founder at Astra Optics, a CUHK spin-off that builds ultra-fast nanoscale 3D printers. An MIT-trained mechanical engineer, he left Boston to turn a professor's Science-published breakthrough into a company across Hong Kong, Shenzhen and Hefei.## Episode SummaryFor years, nanoscale 3D printing sold almost only to university labs, because it was too slow for any factory. Now AI data centers are moving from copper to light, and that move needs hair-width lenses that a fast nanoscale printer can prototype in hours. This episode shows how a professor's lab invention becomes an industrial company, and why the money and the factories for that journey are in mainland China.## Key Points- 🔦 Why AI data centers suddenly need lenses as wide as a human hair- ⚡ The one bet that separates this startup from competitors who perfect materials- 💰 The seed offer from Singapore that was 5-6x lower than the offer from mainland China- 🏛️ How Chinese city governments compete to win deep-tech startups- 🎓 Why mainland investors back the professor, not just the startup- 🏭 Hong Kong vs Shenzhen for hardware: where to put the HQ and where to build- 📦 The sourcing rule from Yiwu: find demand first, supply in China follows## Important Links- You can reach Charudatta Datar at https://www.linkedin.com/in/charudatta-datar-667808a/- Vishesh, your host at https://vishesh.space- Checkout Astra Optics at https://astra-optics.com- Hands hurt while you use the keyboard or mouse? https://tryneo.dev## Chapters0:00 - Introduction1:05 - From MIT to a professor's nanofabrication lab in Hong Kong4:54 - Why professors make a PhD student the CEO of their spin-off7:23 - Why the headquarters moved from Hong Kong to Hefei8:49 - RAISe+ and Hong Kong's professor-led ecosystem10:01 - How two-photon lithography prints a 3D pixel13:49 - The bet on speed over materials15:19 - Co-packaged optics: why AI data centers are moving from copper to light19:49 - Why prototyping, not mass production, is the first market21:30 - Selling to universities first, then to industry22:25 - Funding: the hardest problem in deep tech23:49 - Government grants in Hong Kong, Hefei and the EU25:30 - Why mainland VCs invest in the professor27:50 - Hong Kong vs Shenzhen vs mainland China for hardware30:19 - The Singapore offer that was 5-6x lower32:19 - How China's local governments compete for startups33:49 - An Indian CEO in China: Shenzhen, the Greater Bay Area and high-speed rail36:42 - Language and culture as a foreign founder38:06 - The pre-Series A round and the pitch to Western investors39:51 - Why hardware moves faster in mainland China41:19 - The biggest unlock: narrowing to a few industry use cases43:48 - Marketing on WeChat vs LinkedIn47:08 - How to source and sell in ChinaTopic: Deep Tech Fundraising in China

# How to Fill a Room With 2,000 People on Zero Budget## Guest IntroductionI'm joined by Vasily Alekseenko, Founder of Rare Founders. He started Rare Founders when his own fundraise stalled, has welcomed more than 25,000 founders and investors through its doors, and is now building SoPhy to help people keep the relationships that matter.## Episode SummaryLondon has 200+ startup events in a single ten-day window, and many of them promise investors who never show up. This episode is a playbook for filling a room with zero budget: free venues earned through partnerships, 60-second pitches, and formats built so nobody gets stuck in a dead conversation. Then it goes further: why 1,000 real relationships beat 13,000 LinkedIn connections, and what it takes to measure the difference.## Key Points- 🍺 How a zero-budget founder booked a free venue for 250 people in two minutes- 💦 Why founders get 60 seconds on stage, and a water gun if they go over- 🚪 The format trick that lets you leave any awkward networking conversation- 💸 Why family offices avoid founders, and the event that got them into the room anyway- 🤝 The partnership rule that gets venues, catering and speakers without paying- 📉 Why a big LinkedIn network can hurt you when you ask for an intro- 🤖 How AI raised user expectations so high that a manual MVP no longer works## Important Links- You can reach Vasily Alekseenko at https://www.linkedin.com/in/alekseenkovv/- Vishesh, your host at https://vishesh.space- Check out Rare Founders at https://www.rarefounders.com- Hands hurt while you use the keyboard or mouse? https://tryneo.dev## Chapters0:00 - Introduction0:58 - What Rare Founders is: 25,000+ founders and investors in two and a half years1:47 - The failed fundraise that started it all3:28 - Why paid pitch slots with no investors in the room felt like a scam4:22 - The pub with a no-show organiser and a free venue for 250 people6:37 - Going full-time and launching London's Largest Demo Day8:52 - Why events are hard to monetise, and the 40+ media partners behind them11:35 - Inside open mic pitching: one minute, raised hands and water guns14:57 - Breaks that let you exit an awkward conversation16:28 - When a volunteer copied the whole format17:28 - What San Francisco events get wrong, and right18:23 - Why UK investors avoid founder events23:49 - Family offices vs VCs: who wants to meet whom25:36 - Moving from open mics to technical and sector events28:42 - How the events business pays the bills, and why it is not the moonshot30:52 - The channels that drive traffic: partnerships, content and word of mouth34:08 - The pub lesson: what a real partnership looks like38:23 - The LinkedIn problem: 13,000 connections, 95% strangers40:57 - Social physics: why well-connected founders reach Series B43:07 - SoPhy's AI agent for your inbox, calendar and WhatsApp49:00 - Can you measure trust between two people?55:32 - Where SoPhy is now, and why launch is limited57:14 - SoPhy vs Boardy and Instinct59:31 - How AI raised user expectations and killed the manual MVP1:03:47 - Using Rare Founders events as SoPhy's go-to-market1:05:14 - SoPhy's one-minute pitchTopic: Startup Community Building

# AI's Real Bottleneck Is Not Chips. It Is the 100-Year-Old Grid.## Guest IntroductionI'm joined by Vikhyat Chaudhry, Co-Founder at Buzz Solutions. He's Forbes 30 Under 30 in Energy and just published a book on why the power grid, not chips, is AI's real bottleneck.## Episode SummaryChips get blamed for every AI bottleneck, but the real constraint is a 100-year-old power grid that takes seven to nine years to build a single new transmission line. This episode breaks down why AI data centers now behave like spiky, unpredictable loads utilities can't model ahead of time, and why a 90% accurate AI model is still only 20% of the job of actually fixing the grid.## Key Points- 🔥 Why AI's real bottleneck has nothing to do with chips- ⚡ Why AI data centers made the grid's load growth jump nearly 6x almost overnight- 🌡️ What a thermal camera on a drone can catch before a transformer blows up- 🏆 How a startup beat GE, Siemens, and Schneider Electric by a 58-point accuracy gap- 🤖 The term he coined to fix AI's biggest misconception in the power industry- 💻 Why "Claude generated it" is not an excuse for bad code in production- 📖 What a college drone project measuring Delhi's air quality has to do with any of this## Important Links- You can reach Vikhyat Chaudhry at https://www.linkedin.com/in/vikhyatchaudhry/- Vishesh, your host at https://vishesh.space- Checkout Buzz Solutions at https://www.buzzsolutions.com- Hands hurt while you use the keyboard or mouse? https://tryneo.dev## Chapters0:00 - Introduction0:13 - What Buzz Solutions does for power utilities2:21 - Why Buzz stays software, not drone hardware3:18 - Writing "Powering Intelligence"5:50 - Gigawatt-scale load swings from AI data centers8:40 - Where new data centers are popping up across the US10:01 - What thermal imaging catches before a transformer fails11:14 - Inside the PowerAI multimodal data platform13:23 - Why there's no public dataset for critical infrastructure AI15:15 - Landing the first utility dataset16:38 - The Delhi drone project that measured an AQI over 50017:36 - Beating GE and Siemens for New York Power Authority19:15 - What makes a utility RFP so grueling21:37 - Finding an RFP partner through Google22:19 - Building a partner program that actually works23:38 - The biggest misconception about AI model accuracy25:54 - Rules for AI-generated code on the engineering team27:33 - Scaling code review with CodeRabbit31:45 - Buzz Solutions is hiring, and where to find the bookTopic: AI Grid Inspection

# What This CTO Checks Before He Merges AI Code## Guest IntroductionI'm joined by Michael Bromley, CTO & Co-Founder at Vendure. He started Vendure as a solo open-source project in 2018, and today he no longer writes code: he runs parallel AI agents and a review agent called Nigel.## Episode SummaryCode is no longer scarce, so an engineer's value moves to review, architecture and taste. This episode goes inside a real agent workflow: an orchestrator session that splits Linear issues into parallel lanes, agents that coordinate over Slack, and a one-file review agent that beats a paid tool. It also asks what that shift does to open-source licensing and to the junior engineer's career path.## Key Points- 🤖 Why a single markdown file in a GitHub Action out-reviews CodeRabbit- 🧵 How two engineers stop their agents from colliding: let the agents talk to each other on Slack- 📜 The one thing worth writing in your AGENTS.md (and why most of it is wasted context)- ⚖️ Why an MIT-to-GPLv3 license change made sense in 2024 and makes less sense now- 🎁 The paid feature Vendure gave away for free, and why it is too late to take it back- ✍️ Why a CTO who lets AI write all his code still refuses to let it write his posts- 🎓 What a junior engineer should do when AI does every junior task better## Important Links- You can reach Michael Bromley at https://www.linkedin.com/in/michael-bromley-ab1704156- Vishesh, your host at https://vishesh.space- Checkout Vendure at https://vendure.io- Hands hurt while you use the keyboard or mouse? https://tryneo.dev## Chapters0:00 - Introduction1:33 - What Vendure does for B2B commerce2:13 - From a family art-supply shop to a solo open-source project3:55 - Why Magento, Shopware and WooCommerce were not enough6:41 - From AI skeptic to fully AI-pilled8:22 - Should companies still hire junior engineers?10:38 - The skills that still matter when AI writes the code12:46 - What to put in AGENTS.md to make a codebase agent-ready14:47 - MIT to GPLv3: does licensing matter when code is not scarce?17:58 - How an open-source company makes money19:17 - Michael's dev setup: orchestrators, sub-agents and Linear21:38 - Agents that coordinate with each other over Slack23:20 - Nigel, the review agent that beats CodeRabbit24:58 - The newest change: Codex as the orchestrator28:32 - The channels that bring Vendure customers29:25 - The feature Vendure should have charged for30:52 - Superpower and weakness: taste vs. marketing33:43 - What works on LinkedIn for a developer35:33 - Where to find Michael and Vendure37:11 - Why he still writes his own postsTopic: AI Code Review

# Your AI Agent Does Not Need an API. It Needs a Login.## Guest IntroductionI'm joined by Furqan Kidwai, Co-Founder & CEO of Plouton AI, which builds AI finance co-workers for mid-market CFO teams. A Forbes 30 Under 30 honoree who started his career building credit-risk models at Deloitte, he explains why his agents log in instead of integrating.## Episode SummaryMid-market finance teams do not lack tools. They lose days copying data between them, because most of their ERPs and bank portals have no API. This conversation is about AI agents that log in like a person and finish 10 days of back-office work in an hour, and what it takes to sell automation where every click must be provable.## Key Points- 🔐 Why the API-first automation wave skipped most mid-market finance teams- 🧾 The audit trail that lets an AI agent touch the books without keeping the data- 🤖 How an AI agent gets past MFA prompts the same way a new hire does- 🔀 The outbound pivot that turned a would-be competitor into a sales channel- 🎙️ The webinar playbook that fills a CFO pipeline, and what has to come first- 📉 Why a fintech with 26,000 customers still had to shut down- 🧭 Services as a software: why this AI company will not call itself SaaS## Important Links- You can reach Furqan Kidwai at https://www.linkedin.com/in/furqankarimkidwai- Vishesh, your host at https://vishesh.space- Checkout Plouton AI at https://plouton.co- Hands hurt while you use the keyboard or mouse? https://tryneo.dev## Chapters0:00 - Introduction0:07 - From theatre and computer science to Deloitte credit-risk models1:33 - The grandma pitch: what Plouton AI actually does2:15 - Copilot vs co-executor: agents that finish the work4:47 - Why API-first tools miss mid-market finance teams6:33 - Browser agents in the cloud, like a remote employee7:44 - Handling MFA prompts and session timeouts8:58 - An audit trail without data retention, plus on-prem10:09 - From a 100-person consultancy to services as a software11:22 - The channels that bring in CFO buyers11:50 - The webinar playbook: credibility comes first14:08 - YPay: 26,000 customers, a macro shift, and shutting down16:16 - His founder superpower: staying close to customers17:01 - The pivot to selling through fractional CFO firms18:09 - The founder lesson: stay positive, grow your surface area19:22 - The one product he can't live without20:26 - Where to find Furqan and Plouton AI