AI for Document Q&A with Source Citations by Akhil YadavAI for Document Q&A with Source Citations by Akhil Yadav

AI for Document Q&A with Source Citations

Akhil Yadav

Akhil Yadav

Self-initiated portfolio project. Demonstrates implemented functionality and is not presented as a client engagement. Features and integrations are described according to their verified implementation status.

Overview

DocuLens is a document question-answering application. It extracts text from supported formats, retrieves relevant passages and presents answers alongside source references. It emphasises evidence traceability and clear handling of questions the document cannot answer.
Document workspace
Document workspace

Problem

AI answers about documents are hard to trust when you cannot see where they came from.

Solution

Upload a PDF or text file and ask a question. Every answer links to the exact passage, and the quoted sentence is highlighted in the document viewer.
Answers with citations
Answers with citations

Key features (implemented)

PDF and text upload with size, type and content validation; clear errors for scanned or malformed PDFs (no OCR is claimed)
Passage chunking with page or line numbers; BM25 retrieval with an evidence threshold
"Insufficient evidence" responses instead of guesses
Optional AI answers whose citations are verified against the retrieved text; the live demo uses quote-based answers, labelled as rule-based
Citation inspection with highlighting, copy, Markdown export and clear document

Engineering

In-memory processing (nothing is stored on the server); document text is treated as untrusted data; tests generate real PDFs. 37 unit/integration tests and 10 browser tests (desktop and mobile).

Limitations

No OCR. Keyword retrieval can miss answers phrased differently from the question. Files are limited to 4 MB and 200 pages.
Like this project

Posted Oct 18, 2026

A working document Q&A app: upload a PDF or text file, ask questions, and see the highlighted passage behind every answer, or an "insufficient evidence" reply.

Likes

0

Views

0

Timeline

Oct 18, 2026 - Oct 18, 2026