41 ms
95th-percentile search time over 200K complaints, including encoding the query
results file

Evaluation

30
topics BERTopic found across the 200K complaints; 55% of them (110,456) fit none
results file
6.3×
faster embedding with MiniLM than with BGE: 374.6 against 59.7 texts a second, on 5,000 texts
results file

Timed on 200K complaints on a T4 GPU in Google Colab: 20 fixed queries, five results each, one run, in milliseconds. Vector search includes encoding the query. There are no relevance labels, so this measures speed, not whether the results are right.

Retrieval latency, Complaint Intelligence System
methodp50 (ms)p95 (ms)p95 (ms) as a bar
Vector (FAISS)3541
BM25589929
Hybrid (RRF)614959
Reranked hybrid9111,356
Retrieval latency, Complaint Intelligence Systemresults file

Stack

  • FAISS
  • Sentence-Transformers
  • BERTopic
  • Streamlit

Contact / Rourkela, Odisha, India

Write to me