hjLabs AI Playground
Starting the on-device runtime…
Extractive question answering does something a chat model cannot promise: it returns a span copied verbatim out of your text, with a start offset and a confidence score, so the answer is provably grounded in the source. DistilBERT fine-tuned on SQuAD is small, fast and well suited to pulling clauses out of contracts, policies and documentation. It cannot invent an answer, which is precisely the point.
question answering onlineextractive qa demosquad model browserdistilbert qadocument question answering freeanswer from passagegrounded question answering
A chat model generates text and can hallucinate. An extractive QA model can only return a substring of the passage you gave it, so the answer is always traceable to the source — at the cost of never being able to synthesise across documents.
The model still returns its best span, but with a low score. Use the confidence value as a threshold and treat anything below roughly 0.2 as "not found".
No. The model weights are downloaded from Hugging Face to your browser once, and every inference after that runs on your own GPU through WebGPU. Your text, images and audio are never sent anywhere — hjLabs.in has no inference server and no way to see your input.
Small models do. Recent mobile Chrome and Safari expose WebGPU, but memory is the binding limit — stay under roughly 200 MB of weights on Android, and expect iOS to reclaim the tab sooner. The catalog shows every download size up front so you can pick something that fits.
That depends on the individual model, not on this site. Each model page links to its Hugging Face repository, where the licence is stated — most of the catalog is Apache-2.0 or MIT, a few are research-only. We add nothing on top: hjLabs.in claims no rights over anything you produce here.