Weaviate Podcast

Mathew Jacob, lead author of "Drowning in Documents: Consequences of Scaling Reranker Inference" and now a PhD student in ML systems at the University of Washington, joins the Weaviate Podcast to unpack one of the most surprising results in modern search: cross-encoder rerankers get worse as you give them more documents. The paper began during his Databricks internship, where scaling reranking past roughly 100 documents sent recall@10 plummeting, a result so counterintuitive he assumed it was a bug.The conversation digs into why this happens, reframing rerankers through the lens of boosting, rather than being strictly stronger than first-stage retrievers. Cross-encoders are very good at correcting retriever errors within the distribution they were trained on. Full-scoring experiments over 10,000 randomly sampled documents drive the point home, with BM25 beating state-of-the-art cross-encoders. From there, the discussion moves into phantom hits, cases where wildly irrelevant documents scored highly. For example, a dishwasher document surfacing for a query about disease in Gabonese children. We also discuss whether ensembling rerankers can patch these false positives.The second half explores what comes next for reranking: prompt-based listwise reranking with sliding windows, which proved far more robust than pointwise scoring; RankZephyr-style fine-tuning versus encoding learning signal in prompts with GEPA and DSPy, reasoning rerankers like Rank1 and their latency trade-offs, hard negative mining behind ZeroEntropy's zELO, and pairwise and setwise designs that sit between cross-encoders and full listwise ranking. Adaptive retrieval comes into focus through Natural Language Query to Configuration for Retrieval Agents, predicting per query whether to run simple retrieval, multi-hop, or full agentic search to push the cost-quality frontier.The conversation lands on TraceLab, from Mathew's lab at UW: 40,000 real traces harvested from Claude Code and Codex usage, revealing how coding agents actually behave, prefix cache patterns, long-tailed tool calls, and how understanding these workloads unlocks the next generation of serving optimizations.

Weaviate co-founders Bob van Luijt and Etienne Dilocker return to the Weaviate Podcast to celebrate seven years of building the company, answering questions submitted by the community. The conversation opens with what excites them most in AI right now: Etienne on agentic coding and the "Moore's law" of how long models can sustain autonomous loops, and Bob on world models, new architectures that could slash training energy costs, open source frontier models, and inference on new chips.From there, the discussion dives into taste and the "AI slopification" problem, why AI-generated emails, websites, and decks all look the same, how three job candidates submitted nearly identical AI-built presentations in one week, and why Weaviate runs a dedicated "slop pass" skill over every pull request to strip out phrases like "the smoking gun" and "load-bearing invariant." The human touch, they argue, is now the easiest way to stand out.The Co-Founders then retell their origin story: meeting at a European enterprise company, rewriting a NodeJS prototype in Go, betting on NLP before anyone called it AI, adopting HNSW when it was still a niche paper, and raising a $1.2M seed round from Zeta during COVID. When ChatGPT and the RAG paper hit, Weaviate had a fully working product ready for the wave.Looking forward, Bob breaks down the commoditization playbook that hits every new database category, the same skepticism MongoDB faced, and shares that the number one reason new customers cite for choosing Weaviate is that an LLM recommended it. Etienne makes the case that vector databases are evolving into context engines: context rot is real, stuffing everything into a long context window is inefficient, and retrieval, hybrid search, and structured data all serve one goal, the best possible context. The conversation lands on memory for AI agents, where the hard problem isn't what's worth remembering, but what's worth recalling.

Dr. Bradley Allen brings five decades of AI history into a deep conversation on knowledge engineering, neurosymbolic AI, and the future of enterprise intelligence. The discussion begins with the boom-and-bust cycle of rule-based expert systems, AI winters, and why today’s large language model wave may be different. The conversation then turns to how knowledge is organized in practice, from personal piles of papers searched on demand to formal knowledge graphs built with classes, relations, ontologies, A boxes, T boxes, description logic, and subsumption-based reasoning. Allen explains why semantic web and biomedical ontology successes still leave unresolved questions about cost, maintenance, and whether LLMs can dynamically structure information in ways that preserve meaning. That leads into natural language concept definitions, LLM-based classifiers, rationales, probabilistic reasoning, and the challenge of updating classes as new edge cases emerge.From there, the focus widens to vector databases, semantic search, RAG, topic modeling, distributional semantics, and the ongoing revision required for systems that can never be “once and done.” Allen connects modern LLM behavior to the long history of formal languages, from Frege, Russell, Wittgenstein, Turing, and Gödel to theorem proving, soundness, completeness, paraconsistency, paracompleteness, and the pragmatic tradition of meaning through use. The closing stretch explores world models, reinforcement learning, tool-using agents, enterprise knowledge workflows, role-based access control, governance, normativity, and alignment, ending on the need to build accountable AI systems that channel powerful technology toward responsible outcomes.

Başak Eskili joins the Weaviate Podcast to explore how one of the world’s largest travel platforms adopted vector search, retrieval-augmented generation, and agentic AI at production scale. The conversation begins with Booking.com’s shift from keyword matching to semantic retrieval as internal teams needed embeddings, similarity search, and eventually GenAI RAG workflows. Başak explains why OpenSearch was a practical first step on AWS, how adoption grew across teams, and why hundreds of millions of embeddings, strict latency requirements, complex filtering, and rising concurrency pushed the platform toward Weaviate.The discussion then moves into Booking.com’s partner-to-guest messaging agent, a production GenAI system that helps accommodation partners answer guest questions about check-in, parking, special requests, and reservation details. Başak breaks down the tool-calling architecture, where Weaviate retrieves relevant response templates while GraphQL APIs fetch property and booking context. The agent can suggest templates, craft grounded replies, or decline to answer and leave the conversation to a human, highlighting the practical role of human-in-the-loop design. Evaluation spans offline datasets, LLM-as-a-judge scoring, A/B testing, and live partner feedback.From there, Başak describes the platform engineering behind AI at Booking.com: a central MCP server for internal APIs and external tools, a GenAI gateway for model access, PII reduction, guardrails, prompt injection detection, logging, traceability, and cost tracking across large-scale LLM usage. She also details Booking.com’s evaluation process of Weaviate, including 100 million embeddings, filtered vector search, multi-threaded concurrency testing, reads during writes, and cost-efficient infrastructure provisioning.The episode closes with Başak’s path from computer science and NLP to MLOps and AI platforms, then looks ahead to practical AI, personalized travel agents, and memory systems that capture user preferences, session context, semantic memory, and long-term personalization for future agentic travel experiences.

Dr. Nandan Thakur returns to the Weaviate Podcast fresh off defending his dissertation to discuss the evolution from neural retrieval to agentic search and his new work on Orbit, a synthetic training data pipeline for search agents. The conversation opens with reflections on his PhD journey, tracing the field's shift from ColBERT-style models and sparse retrievers through RAG and into today's agentic search paradigm where LLMs iteratively search, reason, and refine.The discussion dives deep into how Orbit generates multi-hop, riddle-style training queries using DeepSeek's API on a personal laptop over four to six months, making high-quality search agent training data accessible without massive compute budgets. Thakur draws a sharp distinction between deep research (broad, multi-tool report generation) and search agents (focused on search and browse tools to answer specific questions), then connects Orbit's multi-hop queries to BrowseComp's filter-style riddles where each clue narrows the answer space like a funnel. The conversation explores the design of deep research harnesses, chunking strategies, Anthropic's contextual retrieval for entity disambiguation, context compaction to manage bloated agent contexts, and memory services like Weaviate's Engram for compressing search results between reasoning rounds.From there, the episode tackles sequential versus parallel search trajectories, the pass@K approach to rollouts in GRPO training, and whether isolated trajectories should share progress through message passing. Thakur makes a compelling case for training search agents to produce keyword-focused queries optimized for BM25 versus semantic queries for dense retrieval: the idea that one query does not fit all search engines. The conversation closes on future directions: efficiency-focused Pareto frontiers for search agents, long-form report generation evaluation through TREC RAG, and the coming wave of multilingual and multimodal search benchmarks.