Question Answering on SQuAD

A comparative study of extractive question-answering models and standardized evaluation.

01 / Context

The problem

Model comparisons are difficult to interpret when training and evaluation pipelines differ between experiments.

02 / Implementation

Compared BERT, BiDAF, and DistilBERT with a shared training and evaluation workflow across accuracy, precision, recall, and F1.

03 / Current outcome

A benchmark-oriented foundation for later applied retrieval work.

Boundary

No public deployment or repository link is claimed yet.

Continue exploring

Applied AI / Fintech

Evidence-first research workspace for Indian stocks and mutual funds.

Applied ML / RAG

A synthetic-data, non-clinical research sandbox for bounded RAG, citation validation, and RAFT fine-tuning.