Projects

Projects

Selected projects, tools, and research.

Preview of the Ember GPU inference dashboard

Ember

A Kubernetes control plane for GPU inference endpoints.

Core features: Autoscaling, scale-to-zero, cache-aware placement, and monitoring.

Backend stack: Go · Kubernetes · KEDA · Prometheus · vLLM

Live preview ↗Source ↗

English preview of the Pocket48 replay website

Pocket48 Replayer

A replay-to-clip app with transcripts, summaries, and synchronized subtitles.

Core features: AI detects and clips highlight moments; polls live voice rooms and records audio.

Backend stack: Python · FastAPI · FFmpeg · DashScope · Alibaba Cloud OSS

Live preview ↗Source ↗

Preferred Time recommendation pipeline: ALS, HNSW ANN, LambdaRank, and MMR

Preferred Time

An offline-first MovieLens-25M recommender built to compare recommendation quality, diversity, and serving latency.

Recommendation system: ALS and semantic retrieval, HNSW ANN, LambdaRank ranking, and MMR diversity reranking.

Backend engineering: Versioned model serving through FastAPI, Airflow retraining, and optional Kafka-to-S3 event ingestion.

Backend stack: FastAPI · Airflow · Kafka · S3

Source ↗

Publication

SceneBind: Binding What and Where Across Vision, Audio and Language

NeurIPS 2026 Spotlight. Mingfei Chen, Zijun Cui, Ruoke Zhang, Hyeonggon Ryu, Eli Shlizerman.Paper ↗Project page ↗

Tools

Preview of the LeetCode company questions website

LeetCode Company Questions

Organized around 0x3f's topic-based LeetCode problem lists, with difficulty scores built in.

Open website ↗

Rendered sample resume from Resume Versioning Skill

Resume Versioning Skill

A skill for composing and verifying multiple LaTeX resume versions from shared content.

Source ↗