MLOps.community

Jaipal Singh Goud, a Solutions Architect at Prem AI, dives into the exciting world of fine-tuning small language models for personalized AI agents. He discusses the contrast between general LLMs and company-specific models, addressing privacy and data control concerns. Jaipal also explores the complementary roles of fine-tuning and RAG systems in query improvement. He emphasizes the importance of user observation for fine-tuning decision-making patterns and envisions a future with countless personalized models, dynamically chosen for each task.

David Jayatillake, an experienced AI leader and former VP at Cube.dev, delves into the intricacies of semantic layers and their crucial role in data management. He critiques proprietary BI tools for locking companies into confusing ecosystems, advocating for open-source solutions. The discussion extends to how AI agents can streamline data workflows by automating repetitive tasks and enhancing queryability. Jayatillake also highlights the potential of LLMs in building semantic layers and the significance of company-specific definitions for effective data analysis.

Siddharth Bidasaria, a key member of the Claude Code team at Anthropic, shares insights into the innovative coding product's journey. He reveals how Claude Code evolved from a terminal prototype, attracting immediate internal interest. The conversation highlights user-driven improvements like local file tools that enhanced workflow, and the importance of test-driven development for reliable AI code. Siddharth also discusses the balance between model steerability and user friction, plus exciting future possibilities with sub-agents and customizable permissions.

Biswaroop Bhattacharjee, a Senior ML Engineer at Prem AI, dives into the fascinating world of AI memory systems. He discusses Cortex, an innovative framework inspired by human cognition, highlighting how it manages long-term and multimodal memories. The conversation challenges the boundaries of agentic memory, weighing the necessity of forgetting and the implications of memory consolidation. Biswaroop also shares insights into hierarchical collections, retrieval techniques, and the pursuit of integrating vision and audio for a richer AI memory experience.

Marco Palladino, CTO and co-founder of Kong, dives into the complexities of AI infrastructure. He discusses the importance of building AI gateways to enforce governance and security as technology evolves. The conversation touches on the role of agentic workloads and the challenges of MCP servers. Marco speculates on how agents could transform user interactions and even SEO dynamics. He also highlights real-world applications across industries and shares insights on product development strategies. Prepare for an enlightening exploration of AI's future!

A winning team at the hackathon reveals their groundbreaking AI travel agent that manages group trips from start to finish. They share insights on overcoming design challenges and integrating multiple agents. The conversation delves into secure, seamless payment systems without human intervention. User experience is highlighted with interactions via calls and WhatsApp. The use of automation for bookings is impressively detailed, showcasing their rapid traction and team's collaborative spirit. Join them as they invite fellow developers to innovate with Unicorn Mafia!

Varun Khare, Founder and CEO of NimbleEdge, and Neeraj Poddar, Co-founder and CTO, dive into the revolution of on-device AI agents. They discuss why now is the perfect time for this technology, highlighting the hurdles of overmarketing and platform diversity. The duo also explores practical capabilities of on-device models, optimized through lightweight runtimes, and the evolution of personalized multi-agent systems. They delve into the privacy advantages of on-device computing and the potential of AI-first apps to transform user experiences.

Chiara Caratelli, a data scientist at Prosus Group, emphasizes the critical nature of evaluations in AI development. She discusses the importance of stress-testing and building trust through rigorous evaluations rather than merely relying on larger models. Chiara shares her approach to bootstrapping evaluation sets and the role of user feedback in refining these tests. She also touches on the significance of simulating real-world interactions and the need for effective error analysis to enhance agent performance. Learn how her insights can impact the future of AI.

Hannes Mühleisen, co-founder and CEO of DuckDB Labs and Professor of Data Engineering at Radboud University, discusses the innovative DuckLake lakehouse format. He explains how DuckLake transforms data management by separating metadata and computation, enabling a decentralized approach while maintaining centralized control. The conversation covers its rapid adoption due to simplicity, governance models avoiding feature bloat, and surprising community use cases. Hannes also shares insights into upcoming priorities for DuckLake and its potential impact on larger organizations with multiplayer workflows.

Russ d'Sa, CEO of LiveKit, shares his journey from an open-source project to powering voice interfaces for giants like OpenAI. The turning point came when LiveKit collaborated on ChatGPT’s voice features, showcasing the challenges of making AI sound human. He discusses the future of voice in multimodal AI, the importance of minimizing latency for real-time communication, and how entrepreneurial adaptability has shaped their innovative path. Russ believes voice holds rich signals that can transform human-AI interaction.