AskNet-Lite: Distilled Intelligence based Lightweight QA Assistant
0.0 • 0 Reviews

AskNet-Lite: Distilled Intelligence based Lightweight QA Assistant

           The rapid growth of digital documents has increased the need for systems that can retrieve useful information from user-provided content without requiring users to re...

$ 59.00
AvailabilityAvailable Now CategoryPython
CodeFS-010 Specs4 points
Quality Guaranteed Reviewed project files and clear structure.
24/7 Support Reach us anytime for setup and usage help.
Instant Access Download your files right after checkout.
Source Included Code, assets, and project-ready materials included.
Need a custom version?

We can help you adapt this project for your topic, tech stack, or submission requirements.

Talk to us

About This Product

AskNet-Lite: Distilled Intelligence based

Lightweight QA Assistant



1. Introduction

           The rapid growth of digital documents has increased the need for systems that can retrieve useful information from user-provided content without requiring users to read every page manually. Finding a precise answer in lengthy reports, notes, manuals, and other documents can be time-consuming, particularly when the user needs a direct response to a specific question.Question answering (QA) systems address this need by identifying an answer span from a given context in response to a natural-language question. Transformer-based language models have improved extractive QA by learning contextual relationships between questions and passages. However, large language models may require substantial computational resources and can be unnecessary for focused document-based question answering.This project proposes AskNet-Lite, a lightweight document-based QA assistant built with the Django web framework and the distilled DistilBERT model (distilbert-base-cased-distilled-squad). The application accepts PDF, TXT, and DOCX files, extracts their text, and uses the uploaded content as context for answering user questions. The objective is to provide a simple browser-based interface for document-grounded answers with a comparatively compact transformer model.In the current implementation, the pretrained QA model performs inference; the uploaded document is supplied as context and is not used to train or fine-tune the model. The system is therefore an extractive QA assistant rather than a document-specific model-training platform.

2. Problem Statement

           Conventional document review and information lookup involve several practical limitations:

Time-consuming manual search: Users must scan long documents to locate information relevant to a particular question.

Fragmented information: Important details may be distributed across different paragraphs or pages, making manual retrieval difficult.

Technical barriers: Many QA tools require complex setup, specialized interfaces, or high computational resources.

Document compatibility: Users may work with different common formats, such as PDF, plain text, and Word documents.

Context-dependent answers: General-purpose responses may not be grounded in the specific document the user wants to consult.

There is a need for an accessible, lightweight application that accepts common document formats, extracts readable text, and returns answers based on the supplied document context.

3. Input & Dataset Specifications

Input documents: User-uploaded PDF (.pdf), plain-text (.txt), and Microsoft Word (.docx) files.

Question input: A natural-language question entered through the AskNet-Lite web interface.

Document processing: Text is extracted from the uploaded file using format-specific document-reading libraries.

QA context: The extracted document text is passed to the question-answering pipeline as context.

Model: distilbert-base-cased-distilled-squad, a pretrained DistilBERT model fine-tuned for extractive question answering.

Application framework: Django handles web requests, file submission, session state, and rendering of the user interface.

Dataset requirement: The application does not require a fixed image or tabular training dataset. The uploaded document acts as the inference-time context.

The inspected project implementation limits the context supplied to the QA pipeline to the first 4,000 characters. Consequently, answers may not reflect information appearing later in a longer document. The current implementation also maintains question-and-answer history in the user's Django session, associated with the selected uploaded filename.

4. Expected System Output

              After a document is uploaded and a question is submitted, the application is designed to provide:

Extracted-context response: An answer predicted from the supplied document text.

Question and answer history: A session-based record of questions and corresponding answers associated with a document.

Document-specific interaction: The user can ask follow-up questions while using the available document context.

Web interface output: The response is displayed through the Django application interface.

The underlying extractive QA pipeline can return an answer span and confidence score. The current project view reads and displays the answer text; a separately formatted confidence percentage or validated answer-quality score should only be claimed after it is explicitly implemented and tested.

5. Novelty

Lightweight distilled QA: Uses a DistilBERT-based extractive QA checkpoint to support document question answering without building a large model from scratch.

Multi-format document input: Provides a single workflow for PDF, TXT, and DOCX document submission.

Document-grounded interaction: Uses the uploaded document text as the context for answering user questions.

Simple web integration: Combines model inference with a Django interface for file handling and browser-based interaction.

Session-based question history: Organizes questions and answers by uploaded filename within the user's session.

The contribution is primarily the integration and application workflow. The use of DistilBERT itself is based on an established pretrained architecture and checkpoint; claims of a newly invented model architecture or superior performance require separate experimental evidence.

6. Research Contribution

           The project establishes a modular document-question-answering workflow comprising the following components:

Document ingestion: Accepts common document formats through a web upload mechanism.

Text extraction: Extracts textual content from PDF, TXT, and DOCX files using appropriate parsing libraries.

Context preparation: Prepares the extracted text and submitted question for the QA pipeline.

Distilled model inference: Uses the pretrained DistilBERT question-answering pipeline to identify an answer from the supplied context.

Web application development: Uses Django views, URL routing, templates, and sessions to manage the interaction.

Answer presentation: Returns the predicted answer to the user and retains question-answer entries in session history.

For a complete research evaluation, future experiments should report response latency, memory consumption, answer exact match (EM), token-level F1, and performance across a clearly described test set of document-question-answer examples. Results should be measured rather than estimated.

7. Conclusion

              AskNet-Lite presents a lightweight approach to document-based question answering by combining Django with a pretrained DistilBERT extractive QA model. Users can upload PDF, TXT, or DOCX files, submit natural-language questions, and receive answers generated from the extracted document context. The web-based workflow is intended to make focused information retrieval more convenient and accessible.The current implementation provides a foundation for a practical QA assistant. Its effectiveness depends on the quality of extracted text, the relevance of the supplied context, and the model's ability to locate the requested answer. Further work can improve long-document handling through text chunking and answer aggregation, add upload validation and clearer error reporting, and evaluate answer quality and resource usage on a reproducible benchmark. No numerical accuracy or latency claim is made in this document because verified experimental results were not provided.

8. References

[1] V. Sanh, L. Debut, J. Chaumond, and T. Wolf, “DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter,” arXiv:1910.01108, 2019. [Online]. Available: https://arxiv.org/abs/1910.01108

[2] Hugging Face, “distilbert-base-cased-distilled-squad,” Hugging Face Model Hub. [Online]. Available: https://huggingface.co/distilbert/distilbert-base-cased-distilled-squad

[3] Hugging Face, “Question answering,” Transformers documentation. [Online]. Available: https://huggingface.co/docs/transformers/tasks/question_answering

[4] Django Software Foundation, “Django documentation.” [Online]. Available: https://docs.djangoproject.com/

[5] PyPDF2 Contributors, “PyPDF2 documentation.” [Online]. Available: https://pypdf2.readthedocs.io/

[6] python-docx Contributors, “python-docx documentation.” [Online]. Available: https://python-docx.readthedocs.io/


Backend Programming Language

PYTHON

Backend Web Framework

DJANGO

Frontend Structure

HTML

Frontend Styling

CSS

0.0
0 reviews
5
0
4
0
3
0
2
0
1
0

No reviews available for this product.

Add Your Review

View All