Pular para o conteúdo

CTO + Director of AI at Flo Health: Roman Bugaev + Vladislav Nedosekin

2 de março de 2026 · 1 h 42 min

Notas do episódio

In the latest episode, Kyriakos Eleftheriou sat down with Roman Bugaev , CTO of Flo and Vlad Nedosekin Director of AI Platform, at the Terra API HQ in London, to discuss how they built the top health AI platform globally for women's health.

Full episode and show notes: https://tryterra.co/podcast/cto-director-of-ai-at-flo-health-roman-bugaev-vladislav-nedosekin


CHAPTERS


(0:00) Intro — Flo Health: From 20 People to 80 Million Users

(1:02) How Flo became the fastest-growing health company in the world

(1:48) Roman's early days: 20 employees, no revenue, product-market fit

(3:15) How did you know the product was a hit?

(3:25) The underserved women's health market — everyone was building Uber alternatives

(4:31) First ML: neural networks for cycle and symptom prediction

(5:31) Product evolution — from symptom tracking to AI-powered insights

(5:38) Building chatbots inspired by how doctors ask questions

(9:18) A/B testing at scale — Flo's custom experimentation platform

(11:43) Engineering structure: autonomous two-pizza teams

(13:51) Team mistakes — why separate mobile and backend teams failed

(15:29) Scaling from 4 servers to 600 services and petabytes of data

(17:53) "Whenever it's possible, we are NOT doing AI"

(23:15) Why temperature data is critical for ovulation prediction

(25:09) Why Flo is the most accurate period tracker — data diversity advantage

(28:04) Competition: "We don't really have real competitors"

(29:00) AI content creation — generating personalized medical articles

(31:01) Hallucinations vs. conflicting medical sources

(32:34) The three-person blind test: when AI disagrees with humans

(35:10) AI is more consistent than clinicians — but biased against women

(36:52) Fine-tuning open-source models on synthetic women's health data

(38:25) User profile: the foundation of Flo's personalization

(41:19) The digital avatar — your AI health twin that notices what you don't

(43:09) AI router: like a GP triage system for language models

(46:04) Router also controls tone of voice and remembers past conversations

(47:29) "Evaluation, evaluation, evaluation" — how Flo picks models

(48:40) Model stability: why proprietary model updates are dangerous for medical AI

(51:01) Anonymous mode: privacy that enables AI instead of blocking it

(53:49) On-device ML for the most sensitive health data

(56:03) Cloudflare outage — "when everyone is down, you're allowed to be down"

(56:58) Fine-tuning Llama 7B on Databricks — 10,000+ GPU hours per run

(58:07) Training vs. inference cost breakdown

(59:45) 100,000-token prompts: the hidden cost of medical AI

(1:01:05) Build vs. buy: "Build your competitive advantage, buy everything else"

(1:04:47) Value creation vs. value capture teams

(1:07:04) The future: AI that knows you better than you know yourself

(1:09:00) Time series models: the future of health prediction from wearables

(1:10:38) Q&A


Full podcast and writeup → Roman Bugaev and Vladislav Nedosekin on scaling Flo Health to 80M users and shipping AI in femtech

Todos os episódios de Iron Foundry

Configurações

Exibição

Idioma

Define quais estações você vê primeiro na página inicial.

Histórico

As rádios e episódios ouvidos recentemente e suas últimas buscas ficam só neste navegador.

Avisos

Um aviso rápido quando um podcast que você segue tem um episódio novo. Desative podcasts individuais pelo sino na página deles.

Conta

Salve favoritos – no app e no site.

Atalhos de teclado

/
Abrir busca
Espaço
Reproduzir ou pausar
← →
No player: voltar 15 s, avançar 30 s (episódios) · na fileira: próximo bloco
↑ ↓
Na fileira: alvo dentro do bloco
Home / End
Na fileira: primeiro ou último bloco
Esc
Fechar janela ou menu
?
Esta visão geral
Tocando agora

Ir para a estação

A seguir

    Adicione episódios com “A seguir” – eles tocam depois do episódio atual.