ZeroEntropy with Ghita Houir Alami - Weaviate Podcast #150! @Weaviate
ZeroEntropy with Ghita Houir Alami - Weaviate Podcast #150!  @Weaviate
Uploaded November 2026 | Updated November 2026, 1 day ago
Ghita Houir Alami, Co-Founder of ZeroEntropy and now leading Model Research at Notion, joins the Weaviate Podcast to discuss training rerankers, calibrated relevance, and where search models fit in the age of agents. The conversation opens with the ZeroEntropy story: starting from an end-to-end search API, realizing the models powering it mattered most, and shipping a reranker first because swapping in a stateless last-stage API call is far easier for customers than re-embedding a whole corpus. A partnership powering Notion's enterprise search, plus fast query-rewriting models for parallel searches, led to the acquisition.

From there, the discussion dives into small, task-specific decision models and turning cross-encoders into classifiers. Most rerankers trained with triplet loss cluster their scores between 0.4 and 0.6, but a calibrated score unlocks a dynamic K, returning three results or twenty depending on actual relevance, a natural partner to Weaviate's autocut. Ghita then explains zELO, the method behind zerank-1 and zerank-2: hard negative mining eventually surfaces unlabeled positives, so ZeroEntropy instead trained on pairwise comparisons, fit Elo scores through a Thurstone model with sparse sampling of pairwise battles, and distilled a pairwise reranker into an efficient pointwise one that nearly matches its accuracy.

The conversation moves into infrastructure, from quantization-aware training and reranker-specific batching across US GPUs to blending reserved capacity with Modal for bursty traffic. Ghita makes the case for multi-stage retrieval, since embeddings alone hit dimensionality limits, with a small listwise stage on top only when needed. On agentic search, she argues a lightweight reranking step saves expensive LLM tokens, citing Cursor's finding that semantic search made Composer use fewer tokens with fewer undos, while Notion's harness shows that models that search more answer better. Instruction-following rerankers redefine relevance itself: for a memory system, "I'm highly allergic to nuts" is the most relevant answer to "What should I cook for dinner tonight?" Metadata like price is often better handled by a separate model such as XGBoost or Weaviate's Boost API.

The episode closes on the excitement of post-training open-weight models for repeatable tasks inside companies, RSI and using AI to train AI!

Links:
zELO: arxiv.org/abs/2509.12541
zerank-2: https://zeroentropy.dev/articles/zerank-2-advanced-instruction-following-multilingual-reranker/
Score calibration: https://zeroentropy.dev/concepts/score-calibration/
Weaviate autocut: weaviate.io/learn/knowledgecards/autocut
MICE: arxiv.org/abs/2602.16299
zerank-1: https://www.zeroentropy.dev/articles/announcing-zeroentropy-s-first-rerankers-zerank-1-and-zerank-1-small/
Drowning in Documents: arxiv.org/abs/2411.11767
LIMIT: arxiv.org/abs/2508.21038
Should You Use LLMs for Reranking?: https://www.zeroentropy.dev/articles/should-you-use-llms-for-reranking-a-deep-dive-into-pointwise-listwise-and-cross-encoders/
Question's Gambit: arxiv.org/abs/2609.14412
Weaviate Boost: docs.weaviate.io/weaviate/search/boost
Jev as a Reranker (Hindsight): hindsight.vectorize.io/blog/2026/09/24/adding-jev-reranker-what-we-learned
Jev reranker benchmark (kapa.ai): kapa.ai/blog/jev-reranker-benchmark

Chapters:
00:00 Welcome Ghita!
00:52 Founding Story of ZeroEntropy
04:31 Model Research at Notion
09:21 Jev as a Reranker
12:09 Dynamic K for Search
17:14 Beyond Pointwise Comparisons
21:00 zELO
29:30 Infra for Rerankers
36:51 Drowning in Documents
42:39 Agentic Search and Reranking
48:04 Instruction-Following Rerankers
52:16 Metadata in Reranking
54:29 RSI and SLMs
ZeroEntropy with Ghita Houir Alami - Weaviate Podcast #150!What is HNSW?MUVERA with Rajesh Jayaram and Roberto Esposito - Weaviate Podcast #123!Booking.coms Partner-to-Guest Messaging AgentRAGKit with Kyle Davis - Weaviate Podcast #93!Box AI with Ben Kus and Bob van Luijt - Weaviate Podcast #120!Subjectivity in AI with Dan Shipper: AI-Native Databases #4Founding Weaviate with Bob van Luijt and Etienne Dilocker - Weaviate Podcast #140!Scaling Pandas with Devin Petersohn - Weaviate Podcast #101!Generate multimodal datasets for your demos, Proof of Concept (PoC), or project with Generator9000!Think Deepseek-R1 was trained like other top LLMs? Think again.AI in Education with Rose E. Wang - Weaviate Podcast #106!
Weaviate vector database |

ZeroEntropy with Ghita Houir Alami - Weaviate Podcast #150!

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER