A dusk forest grove with kadamba trees, a river, and a peacock on the bank.
Priyank Jayraj

A personal journal

Priyank Jayraj

PhD research scholar · BITS Pilani

I work on information retrieval and applied machine learning. The systems I care about have to hold up under real constraints: latency budgets, messy documents, and languages most models barely see.

On the bench

Current work

Welfare-scheme documents turned into retrieval systems that low-resource language communities can actually use, culturally faithful, not just translated.

Retrieval · graphs

CTU-FlowRAG

Role-conditioned retrieval over typed sentence graphs, so a welfare document can be queried by the role the reader cares about.

0.82 nDCG@10 · 3,048 documents

Culture · RAG

QCCAF

Query-conditioned fusion that adds local cultural context before image prompts are written. Zero-shot across related languages.

0.93 nDCG@10 on Santali

Latency · bandits

Bandit-Quota

A latency-budgeted contextual bandit for per-query retriever selection. Match an ensemble when you can afford it.

2.4x to 9.3x faster than full ensemble

The full research page · Accepted papers