feat: finalize hybrid search pipeline with query expansion and RRF

[sam131725]

Overview

This PR contains the final search results for the DevRev Search Benchmark. The pipeline has been optimized for high Recall@10 and MRR using a multi-stage retrieval strategy.

Key Components

• Query Expansion: Utilized llama3.2:3b to expand user queries for better semantic matching.
• Hybrid Retrieval: Combined Keyword (BM25) and Semantic (qwen3-embedding) scores.
• Fusion: Integrated results using Reciprocal Rank Fusion (RRF).
• Re-ranking: Final Top-10 selection via ms-marco-MiniLM Cross-Encoder.

Files Included

• test_queries_results.json: Official submission file.

work-item: ISS-1

[sam131725]

re-run validation

[prakhar7651]

Do you have a file to be evaluated?

[sam131725]

Please evaluate test_queries_results.json

[prakhar7651]

Hey!
These are your scores.
Recall@10: 0.3156
Precision@10: 0.2154