Mitesh Agrawal has a background in Mechanical Engineering. He was one of the co-founders of Lambda, a company in the supercomputing space, where he spent 8.5 years working on everything under the sun. He's very grateful to be in an industry that is booming, but also aligns with his personal interests. Outside of tech, he is married to an ultra supportive wife, and is enjoying being a new father. He enjoys playing tennis, when he can find time to get to the court, and enjoys a good sci-fi book. He mentioned the Foundation series was one of his favorites, but admits it changes depending on the season.
In 2023, the officers at Mitesh's current venture noticed all of the advancements of AI - in particular, model sizes getting larger. What they realized was that when it comes to inference, memory capacity quickly became a problem... and with this, he and the team got excited about building a new architecture to make it better.
Ryan welcomes Kari Briski, NVIDIA’s VP of Generative AI Software for Enterprise, to the show to explore how a chip manufacturer got into the model development game. They discuss NVIDIA’s co-design feedback loop between model builders and hardware architects, share insights on precision model training and memory management systems, and take a look at the roadmap and development of NVIDIA’s fully open-source Nemotron.
Episode notes:
Nemotron is a family of open models with open weights, training data, and recipes for building specialized AI agents.You can learn more on their Hugging Face page or at NVIDIA GTC on March 16-19.
Show: @pythonbytes@fosstodon.org / @pythonbytes.fm (bsky)
Join us on YouTube at pythonbytes.fm/live to be part of the audience. Usually Monday at 11am PT. Older video versions available there too.
Finally, if you want an artisanal, hand-crafted digest of every week of the show notes in email form? Add your name and email to our friends of the show list, we'll never share it.
cattrs also focuses on functional composition and not coupling your data model to its serialization and validation rules.
When you’re handed unstructured data (by your network, file system, database, …), cattrs helps to convert this data into trustworthy structured data.
Batteries Included: cattrs comes with pre-configured converters for a number of serialization libraries, including JSON (standard library, orjson, UltraJSON), msgpack, cbor2, bson, PyYAML, tomlkit and msgspec (supports only JSON at this time).
“I teach a couple of introductory Python courses and I've been thinking about which advice to give to my students, that are studying how to program for the first time. I have collected my ideas in these blog posts”
Why learning to program is as useful as ever, even with powerful AI tools available.
How to use AI as a tutor rather than a shortcut, and why practice remains the key to real understanding.
What the real learning objectives are: mental models, managing complexity, and thinking like a software developer.
Seyi Ebenezer didn't come to fintech from a hackathon or an accelerator. He came from KPMG's audit desks and Access Bank's corporate finance floors. These are environments where the numbers had to add up before anyone was allowed to dream out loud. That training shows in everything about how he has built Payaza Africa, from claims of launching profitably with a single gas station client to rejecting six or seven VC approaches in favour of bootstrapping a business he could defend on paper.
In conversation with Andile Masuku, Ebenezer — who co-founded Payaza in 2020 and launched in March 2022 — lays out a philosophy that cuts against the grain of Africa's startup narrative. Where the dominant playbook says raise fast, grow faster, and worry about unit economics later, Ebenezer argues that African founders face a structural reality that makes that approach uniquely dangerous: a "natural prejudice rating" on the continent that means even Aliko Dangote isn't immune to credit downgrades.
His conclusion: if the system is stacked against you, your books had better be immaculate.
The conversation covers Payaza's origins solving payment reconciliation for Nigerian fuel stations, why Ebenezer treats every product that isn't profitable within six months as a candidate for shutdown, and how securing investment-grade credit ratings from Augusto & Co, DataPro, and GCR (with a Moody's rating to boot) has transformed the company from price taker to price giver in investor conversations.
Along the way, Ebenezer draws a direct line from the 2008 financial crisis to the recent VC funding winter in African tech, and argues that the founders who built structure survived both.
But the conversation's most striking moment comes near the end with Ebenezer's call for the creation of a pan-African credit rating agency; one that uses community-based risk models suited to how African business actually works, rather than importing Western frameworks wholesale.
Key insights:
On why debt creates discipline: Ebenezer's central thesis is that debt financing forces founders to confront profitability from day one. Unlike equity, where capital can mask weak fundamentals, debt has interest that "does not sleep on Saturday, does not sleep on Sunday." He argues this constraint is a feature, not a bug, particularly for African founders who face structural disadvantages in how the market perceives their businesses.
On building from the books outward: At Payaza, corporate governance came before scale. Ebenezer engaged Deloitte as an auditor from the company's earliest days. It's a decision he says he initially regretted when the first audit surfaced over sixty exceptions. But those painful early investments in structure are what enabled Payaza to access capital markets, raise commercial paper without collateral, and achieve investment-grade credit ratings — outcomes virtually unheard of for a Nigerian fintech.
On the "prejudice rating" African businesses carry: Ebenezer points to World Bank data showing that Africa's default rate on infrastructure funding is just 1.9 per cent (second only to the Middle East at 0.9 per cent) while Western Europe sits at 9.1 per cent. Yet a business headquartered in Western Europe would still receive a higher credit rating. His response: African founders must over-prepare, building the kind of documentation and governance that neutralises bias before they walk into any room.
On rejecting the VC playbook — without rejecting VC: Ebenezer is careful not to demonise venture capital. His argument is about sequencing: build structure first, demonstrate profitability, then engage equity investors from a position of strength. He turned down six or seven approaches during the VC boom, telling his team to trust the longer game. The result: when he now sits across from potential investors, he sets the terms. "Evidence dominates argument," he says.
On why African businesses can't emulate Amazon's playbook: When pressed on whether his conservative approach stifles ambition, Ebenezer invokes the Dangote example. If Fitch can withdraw the credit rating of Africa's wealthiest industrialist, and downgrade Afrexim Bank, then no African founder can afford to assume the market will extend them the patience it gave Jeff Bezos. "If they could touch Dangote," he asks, "who are we?"
On Payaza's efficiency-first growth model: Rather than competing on price — a "race to the bottom" — Payaza competed on settlement speed, offering same-day payouts to merchants using its own capital while competitors operated on T+1 or T+2 cycles. This earned trust and referrals, creating organic growth with thin but real margins. Every merchant is evaluated against an activity-based costing model: if onboarding them isn't profitable, the relationship doesn't proceed.
Notable moments: 1. The Petrocam origin story: Payaza's first client was Petrocam, a Nigerian fuel retailer with 57 filling stations. The problem: reconciliation chaos and shrinkage across distributed locations. Payaza built "Branches," a product that gave the group CFO a centralised, real-time view of collections across every station — eliminating accounting discrepancies, reducing theft, and cutting the finance headcount needed at each site. The product was profitable from day one. "We are solving a problem for them and then we're charging them fairly," Ebenezer recalls. That first deal set the template for everything that followed.
2. The credit rating upgrade that broke the rules: After raising commercial paper on the Nigerian capital market and making an early repayment, Payaza received a credit rating upgrade from BBB- to BBB+ in a matter of months. The norm is a 24-month cycle between upgrades. The rating agency told them they had "a very good case" — a vindication, Ebenezer argues, of prioritising fundamentals over flash.
3. The SME Tribe experiment yielding zero bad debt: When Instagram went down for several days, Payaza saw an opportunity. It built SME Tribe, a web-based marketplace that mirrored what small traders were selling on Instagram, then layered on "Payaza Boost": uncollateralised working capital advances of 25 per cent of a merchant's three-month average collections. The result: zero non-performing loans. Ebenezer uses this as evidence that African credit risk models need to account for community-based accountability, not Western-style board structures.
4. The pan-African credit rating pitch: In the episode's most charged exchange, Ebenezer pivots from discussing his own business to issuing a direct challenge: Africa needs its own credit rating infrastructure, potentially housed under Afrexim Bank or the African Union's APRM framework. He argues that the global rating oligopoly (agencies built "200 or 400 years ago" that keep acquiring regional competitors) cannot adequately assess African risk because Africa is "community-based." His proposed model would incorporate social accountability mechanisms alongside financial metrics. And then, live on the podcast, he nominates Andile Masuku to lead the convening.
Ranjan Roy from Margins is back for our weekly discussion of the latest tech news. We cover: 1) OpenAI hits $25 billion ARR, Anthropic hits $19 billion ARR 2) Are ARR numbers trustworthy? 3) OpenAI's insane revenue expectations 4) Did Apple actually play this perfectly? 5) We need a Tim Cook with claw hands Apple ad 6) AI lab IPOs are brewing, what will the S-1s look like? 7) Anthropic's still talking with the Pentagon 8) Dario's internal memo 9) Wait, was this actually marketing for Anthropic? 10) Or was it a real worry about AI-enabled surveillance? 11) McDonald's CEO's unwitting viral moment
---
Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice.
Want a discount for Big Technology on Substack + Discord? Here’s 25% off for the first year: https://www.bigtechnology.com/subscribe?coupon=0843016b
You're adding type hints to your Python code, your editor is happy, autocomplete is working great. But then you switch tools and suddenly there are red squiggles everywhere. Who decides what a float annotation actually means? Or whether passing None where an int is expected should be an error? It turns out there's a five-person council dedicated to exactly these questions -- and two brand-new Rust-based type checkers are raising the bar. On this episode, I sit down with three members of the Python Typing Council -- Jelle Zijlstra, Rebecca Chen, and Carl Meyer -- to learn how the type system is governed, where the spec and the type checkers agree and disagree, and get the council's official advice on how much typing is just enough.
Ryan chats with Kevin Peterson, CTO of Bedrock Robotics, about the evolution of self-driving technology and why robotics is now advancing; how real data is still relevant but simulation becomes essential for scale; and the future of robotics in addressing labor shortages and enhancing productivity.
Episode notes:
Bedrock Robotics creates technology that upgrades existing heavy equipment, enabling autonomous operation for construction machinery.
Ashwin Agrawal came to the US when he was 17, to Rochester for school. He now lives in the Bay Area, and admits he misses his friends on the east coast, as they all stayed back in that area - but he does NOT miss the winters. He has been building his current venture for 3-4 years, and prior to that, he was as at Google for a decade, apart of Google Cloud's huge growth trajectory. Outside of tech, he has a family with 2 middle school sons, with whom he likes to spend a lot of time with, hiking or eating good sushi.
Ashwin was laid off from a few jobs in the past. After experiencing this, he vowed to build a solution that would help people going through this sort of experience. After the last layoff, he formed his company at 4:30 am in the morning, to help anyone in point A wanting to go to point B.
Michael Horowitz is the former deputy assistant secretary of defense for force development and emerging capabilities at the Department of Defense, and currently a professor at the University of Pennsylvania. Horowitz joins Big Technology to discuss the Anthropic–Pentagon rupture and what it signals about how the U.S. government wants to use frontier AI. Tune in to hear his inside view on how models like Claude actually get deployed in defense workflows, why a contract fight over “mass surveillance” language escalated, and what the trust breakdown says about the future of AI partnerships with the state. We also cover autonomous weapon systems vs. “fully autonomous weapons,” what today’s AI can and can’t do on the battlefield, and how AI is likely to reshape warfare over time. Hit play for a clear-eyed look at where Silicon Valley and the national security establishment collide—and what happens next.
---
Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice.
Want a discount for Big Technology on Substack + Discord? Here’s 25% off for the first year: https://www.bigtechnology.com/subscribe?coupon=0843016b
Today, we are continuing our series, entitled Developer Chats - hearing from the large scale system builders themselves.
In this episode, we are talking with Oleksandr Piekhota, Principal Software Engineer at Teaching Strategies. Oleksandr helps to show us at what point of scale platform approaches are required, when to run experiments and when to stop, and perhaps more importantly - engineering ownership beyond the code.
Questions
You’ve moved from hands-on engineering into principal and technical leadership roles, working on architecture and platforms.At what point did you realize your work was no longer about individual features, but about the system as a whole
Across several projects, growth didn’t break functionality — it exposed architectural limits.Can you recall a moment when it became clear that shipping more features wouldn’t solve the problem, and a platform approach was required?
You’ve designed and supported APIs end-to-end, from architecture to real customers. How do you distinguish between an API that simply works and one that can truly support business scale?
Internal systems like invoicing and HR workflows began as automation, but evolved into real products.What tells you that an internal tool is worth developing seriously rather than treating as a temporary workaround?
In R&D, you explored CI/CD automation, server-less, and infrastructure experiments — not all reached production. How do you decide when an experiment should continue, and when it’s no longer worth the engineering cost?
You’ve hired teams, set standards, and shaped long-term technical direction. At what point does an engineer stop being a contributor and start owning business-level outcomes?
You contributed to open-source tools that later became part of your company’s infrastructure. Why do you see open source contributions as part of serious engineering work rather than a side activity?
Looking across your projects, how do you now recognize a truly mature engineering system? Is it code quality, process, or how teams respond when things go wrong?
If we look five to seven years into the future, which architectural assumptions we treat as “standard” today are most likely to turn out to be naive or limiting?
