Grad student at UESTC (Chengdu) · retrieval-augmented generation & retrieval systems.
Good generation starts with good retrieval. I work on the pieces that put the right context in front of an LLM — and prove it did.
The pipeline I build toward
query → [dense retrieval] → [rerank] → [grounded, cited generation]
Projects
-
densekit — the retrieval half Bi-encoder training, ANN indexing (IVF / LSH / PQ), and IR evaluation on a NumPy core. Offline-first.
-
ragcite — the generation half A pure-Python RAG pipeline with grounded, citation-aware generation and an evaluation harness.
Everything runs offline, no GPU required — NumPy core with an optional PyTorch backend, CI green.