Aryan Mehndiratta

Research Assistant with Retrieval


2026
completed

Research Assistant with Retrieval

A citation-grounded assistant that answers questions over thousands of papers.

Year
2026
Category
LLM Applications
Stack
Python, LangChain, pgvector, OpenAI, FastAPI

The Problem

Describe the problem, who it was for, and why existing tools fell short.

Image
Screenshot of the interface.

Approach

Explain architecture, models chosen and trade-offs.

python
def retrieve(query: str, k: int = 8):
    dense = vector_store.search(embed(query), k=k)
    sparse = bm25.search(query, k=k)
    return rerank(query, dense + sparse)[:k]
Hybrid retrieval, simplified.
Media 01Image · Video · Audio · PDF · Slides · Document · Any file
01 / 04
Demo gallery — videos, screenshots, slides or documents.

Results

  • Answer accuracy up 32% over baseline.
  • Median latency under 1.2s.
GitHubSource code →Repository and documentation.