137 Podcast

šŸ”Ž LLMs in search. BERT, embeddings, and production reality

1 h 31 min Ā· 12. juli 2025
episode šŸ”Ž LLMs in search. BERT, embeddings, and production reality cover

Description

* How modern search is built and where LLMs fit in. * How semantic search architectures are designed and evaluated * Why BERT is still the baseline in many systems * Fine-tuning strategies, training data, and infrastructure trade-offs * What's possible in multimodal, and what usually isn't worth the effort * Choosing between PostgreSQL and vector databases * What often blocks companies from improving search, from training data to infra 🄐 Hosts - Anton Zhuravsky [https://www.linkedin.com/in/anton-zhuravsky-40690124/] & Dmitry Volokh [https://www.linkedin.com/in/davolokh/] ā˜•ļø Guest – Roman Grebennikov [https://www.linkedin.com/in/romangrebennikov/], who has been building search at Delivery Hero, and has also built his own search engine – Nixieserach.ai The podcast is brought to you by ⁠On The Spot Development⁠ [https://onthespotdev.com/], a software development company which Anton founded. They build R&D teams for startups and tech companies.

Comments

0

Be the first to comment

Sign up now and become a member of the 137 Podcast community!

Get Started

2 months for 19Ā kr.

Then 99Ā kr. / month Ā· Cancel anytime.

  • Podcasts kun pĆ„ Podimo
  • 20 lydbogstimer pr. mĆ„ned
  • Gratis podcasts

All episodes

5 episodes

episode How AI is changing engineering teams and companies artwork

How AI is changing engineering teams and companies

In this episode, we talk to Boaz Adato (Head of Engineering Guild) about how AI is changing engineering, the uneven impact of AI across new and legacy systems, and how workflows and team autonomy are evolving in practice. šŸ” In the episode: * How AI adoption started (and why many engineers rejected it at first) * What actually changed inside teams,. Workflows, roles, and ownership * Why AI works great in greenfield projects but struggles with large codebases * How non-engineers (finance) started building their own tools * Why key engineering metrics (like PR throughput) haven’t improved as expected * Wild guesses for 2026-2027 šŸ§‘ā€šŸ’¼Guest: Boaz Adato [https://www.linkedin.com/in/boaz-adato-9a89562/] šŸŽ™ļøHosts: Anton Zhuravsky⁠ [https://www.linkedin.com/in/anton-zhuravsky-40690124 ] & ⁠Dmitry Volokh [https://www.linkedin.com/in/davolokh ] šŸ“137 Podcast on X: https://x.com/137_podcast [https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqblJ4OFpNRFZDUVlaLW11R2VuVTVQMDVsc2phZ3xBQ3Jtc0tsbEg2QmI4emxRTGFqcWV6Z2UxR3ZBQ25INkF4bE9QOEpZaUFWaUVKeVNBYUtmQ3paM2RST09pWGxzUUdoVjI2SDhGaXgyOFduUzhYLS1idjM5RThsemM1TUdVRFR2bTI4VlRUNFdndkpCZmJLREdZcw&q=https%3A%2F%2Fx.com%2F137_podcast&v=SP4QqyOBmdg] šŸ™Œ This podcast is brought to you by On The Spot Development. [https://onthespotdev.com/]We help tech companies build R&D teams in Poland and have open roles for engineers. Check out the relevant opportunities for AI Engineers: https://onthespotdev.com/careers?*=AI [https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqa1B1ZzBZdFF3WU44dUhjbWFBdmE0Vk1yblBYUXxBQ3Jtc0ttR3VndU9JRng4ZzNfSEFiVWxlYXFKMmxTOEpvaFpaVVRNb1ZMVWFZTmY3U2RMOEhseHc5X3RJcEdQaFN6ZURlZEhpSlQ3RnJfLTFhdWVQOURlTkE2U2F0SUpnTlJJemRlSzVnX2IyN3FvelFVNVowdw&q=https%3A%2F%2Fonthespotdev.com%2Fcareers%3F%2A%3DAI&v=SP4QqyOBmdg]

7. maj 20261 h 0 min
episode Building AI apps. Security issues, UX, and future of software engineers artwork

Building AI apps. Security issues, UX, and future of software engineers

* What changes when you move from API demos to production apps (rate limits, multi-cloud, failover strategies) * Security challenges: prompt injection, MCP as remote code execution, supply chain risks, least-privilege agents * How to measure reliability (confidence in AI responses, cost-of-error, online and offline evals) * Why ā€œcode is cheapā€ now and how that’s reshaping engineering roles * "Trustworthy" UX patterns: transparent agent steps, generative UI, ambient agents * The job shifts already happening: juniors struggling to get hired, first-line support and QA under threat * How AI is lowering the barrier to entry in education and what engineers need to stay relevant šŸ§‘ā€šŸ’¼Guest: Eden Marco [https://www.linkedin.com/in/eden-marco/ ] šŸŽ™ļøHosts: Anton Zhuravsky [https://www.linkedin.com/in/anton-zhuravsky-40690124 ] & Dmitry Volokh [https://www.linkedin.com/in/davolokh ] The podcast is brought to you by ⁠⁠On The Spot Development⁠⁠ [https://onthespotdev.com/], a software development company which Anton founded. They build R&D teams for startups and tech companies.

18. dec. 20251 h 14 min
episode šŸ”Ž LLMs in search. BERT, embeddings, and production reality artwork

šŸ”Ž LLMs in search. BERT, embeddings, and production reality

* How modern search is built and where LLMs fit in. * How semantic search architectures are designed and evaluated * Why BERT is still the baseline in many systems * Fine-tuning strategies, training data, and infrastructure trade-offs * What's possible in multimodal, and what usually isn't worth the effort * Choosing between PostgreSQL and vector databases * What often blocks companies from improving search, from training data to infra 🄐 Hosts - Anton Zhuravsky [https://www.linkedin.com/in/anton-zhuravsky-40690124/] & Dmitry Volokh [https://www.linkedin.com/in/davolokh/] ā˜•ļø Guest – Roman Grebennikov [https://www.linkedin.com/in/romangrebennikov/], who has been building search at Delivery Hero, and has also built his own search engine – Nixieserach.ai The podcast is brought to you by ⁠On The Spot Development⁠ [https://onthespotdev.com/], a software development company which Anton founded. They build R&D teams for startups and tech companies.

12. juli 20251 h 31 min
episode Engineering Culture šŸš€ The science of hiring, firing and running 400 experiments per quarter artwork

Engineering Culture šŸš€ The science of hiring, firing and running 400 experiments per quarter

How to build & maintain an engineering culture of experimentation and resilience? We sit down with Roman, CTO of Flo Health, and discuss how to build a high-performing tech culture, including: šŸš€ How to win the "best invention" award as a result of engineering response to a privacy challenge. šŸ› ļø The hiring playbook that brings in 10 engineers in a day (and actually works). šŸ”„ "Oh no" moments: Handling media storms and sudden relocations. šŸ“Š The science of running 400 experiments per quarter (without setting everything on fire). šŸ¤– Future: $200kk in RnD, AI (obviously), wearables, and the fine-tuning of everything.Also there – how to hire top engineers, optimize infrastructure costs, and maintain a data-driven culture. Hosts - Anton Zhuravsky & Dmitry Volokh, Guest - Roman Bugaev The podcast is brought to you by On The Spot Development [https://onthespotdev.com/], a software development company which Anton founded. They build R&D teams for startups and tech companies.

20. feb. 202555 min