Generative AI Expert Consultation

DEFINITION OF CHALLENGE/PROBLEM TO BE RESOLVED

It is aimed to leverage the potential of NLP and Large Language Models in extracting intelligent insights by linking large datasets, including reports, documents, images, and videos, available with the Client. The solution should operate on standalone hardware without internet connectivity and support a cloud architecture for at least 1000 concurrent users. The architecture must address Big Data's Velocity, Volume, Value, Variety, and Veracity while ensuring security compliance.

(a) Model will be trained on large amount of variety of data such as documents, handwritten text, excel sheets, power point PPTs, word documents, BRs, equipment manuals, IR images, reports, letters, faxes, flow diagrams, charts, block diagrams, videos, training books, medical reports (all variants of medical reports, data and scanned copies).

(b) The format of documents, manuals, reports, letters, faxes will not be restricted to readable PDF format and the solution should cater for all possible formats. Accordingly, solution provider should integrate with the solution a well-developed OCR solution to extract information from scanned documents, handwritten documents and non-readable PDF formats.

(c) Similarly, solution developed should have a capability of extracting information from all possible variants of image and video formats including some of the proprietary formats of images and data

(d) The model should intelligently understand the context of query of the user and should be able to intelligently link information available in textual, image, video or any other format and subsequently provide explanation and insights to the user.

(e) Application / model should be able to display the data to a user in an innovative 2D and 3D visualization format such as Pie-chart, bar graph, time series, line chart, histograms, scatter plot, bar chart, column chart etc, where ever it is feasible and relevant historical data is available.

DETAILED QUALITATIVE REQUIREMENTS (QRS)

Key Features/Functions/Capabilities Required in Proposed Product/Solution

Train on diverse data formats: documents, handwritten text, images, videos, and scanned PDFs, among others.

(b) Integrate an advanced OCR system for extracting information from non-readable formats.

(c) Extract data from proprietary formats of equipment.

(d) Provide contextually relevant responses linking data across formats

(e) Secure user authentication with unique IDs and passwords.

(f) User-friendly interface with session-based query context retention.

(g) Display data innovatively in 2D/3D visualizations.

(h) Accept keyboard and audio input queries.

(i) Generate context-aware responses incorporating diagrams and flowcharts.

(j) Include prompt engineering for improved accuracy.

(k) Enable reinforcement learning and user feedback integration.

(l) Maintain session history and timestamps for entries.

(m) Train models on additional client-provided data.(q) Ensure comprehensive documentation and training materials.

Operational Functionalities

(a) Generate human-like responses to natural language queries.

(b) Maintain conversation context for coherent interaction.

(c) Handle ambiguous queries effectively.(d) Comply with LOCAL systems for SSO and CERT-in certification.

(e) Hardware to support reasonable response times for up to 1000 concurrent users.

(f) Average accuracy above 90%, benchmarked against ChatGPT/Copilot. Responses must be grammatically correct with proper punctuation.(c) Maintain complete session archives with retrieval filters.

(g) Conduct trials per approved QAP/ATP.(b) Submit proposed QAP/ATP for client approval.

Umělá inteligence Generative AI Machine Learning (ML) Multimodal Large Language Model Retrieval-Augemented Generation

Identifikační číslo projektu: #39011704

O projektu