Experimenting with the benefits and downsides of using an LLM alongside BM25 for product search relevance.

An implementation of fair ranking algorithms to ensure equitable exposure and relevance in transaction processing and search results.

Official code for the paper "A New Role for Relevance: Guiding Corpus Interaction in Agentic Search"

Engine-agnostic search-relevance core (instruction rerank steering, LLM category-match, query rewrite, tiered escalation, eval metrics) on @papercusp/rerank. Shared by Restart + papercup.

Offline search relevance judging harness

Data fusion for heterogeneous treatment effect estimation with multi-task Gaussian Processes

Multimodal multi-task short-video recommendation with PLE/MMoE/SharedBottom architectures, uncertainty weighting, and PCGrad optimization. 48+ experiments on THU-ShortVideo dataset with Qwen LLM em…

DCRMTA: Deep Causal Representation for Multi-Touch Attribution

find a segment that has length k among plenty of points by distribute sql within O(n) time complexity

PriCoRec research code, accepted at the 20th ACM Conference on Recommender Systems (RecSys 2026).

HOAM A Hybrid Orthogonal Attention Network for High-Precision, High-Efficiency Industrial Visual Search

Industrial-grade implementation of seq2seq algorithm based on Pytorch, integrated beam search algorithm.

Controllable Multi-Objective Re-ranking with Policy Hypernetworks (KDD 2023)

Multi-objective marketplace ranking engine balancing revenue, conversion, and new-provider cold-start.

RLProp — results-level proportional selection for multi-objective recommender systems: pick top-k items so relevance, diversity, novelty & fairness appear in the ratios you set, via mandate-allocat…

CALISSTA: calibrated recommendations at candidate-retrieval time via sparse-autoencoder steering + D'Hondt aggregation. Model- and dataset-agnostic (bring your own RS backbone/SAE, genres, tags, ca…

📊 A specialized mathematical desktop suite designed for multi-objective optimization and ranking alternatives through group utility and individual regret analysis. 🔐

A C/C++ header-only library for generating text embeddings, sparse embeddings, image embeddings, and document reranking using ONNX Runtime

Implementation of paper: HLATR: Enhance Multi-stage Text Retrieval with Hybrid List Aware Transformer Reranking

**Knowledge-Distillation-for-RecSys-Efficiency:** Explored **Knowledge Distillation** for recommendation systems, achieving **49.5% model compression** (97.5K→49.5K params). Observed **Negative Kno…

Ad studio + campaign engine: Higgsfield creative generation, Facebook-faithful previews, one-click launch, guardrailed optimization

Code for paper "RRC: Unlocking Generative Reward Models in LLM Reinforcement Learning via Ranking-Based Reward Construction"

MARS is shortened for Multi-Agent Research Studio, a library for mulit-agent reinforcement learning research.

DreamX-World: A General-Purpose Interactive World Model

The code of paper Learning Cut Selection for Mixed-Integer Linear Programming via Hierarchical Sequence Model. Zhihai Wang, Xijun Li, Jie Wang*, Yufei Kuang, Mingxuan Yuan, Jia Zeng, Yongdong Zhan…

2021年最新总结 500个常用数据结构，算法，算法导论，面试常用，大厂高级工程师整理总结

Code for KuaiLive-M3

Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1