Akshay Pachaar (@akshay-pachaar)

How Akshay thinks

Evidence‑first retrieval
Prioritizes judging each retrieved passage for evidence strength before it reaches the LLM.
For you: You should filter passages for strong evidence before generation.
Broad candidate pool
Merges dense embeddings and keyword matches with reciprocal rank fusion to retrieve a wide set of candidates.
For you: You can boost recall by combining multiple retrieval methods.
Threshold‑driven filtering
Applies calibrated probability thresholds to decide which passages continue to the generation step.
For you: You can tune a probability cut‑off to control the model’s input.