AI Based Data Matching Tool

I own a ERP tool where we create various reports to pull the data from the ERP system. I have definitions of the reports for about 5000 reports containing their columns, their paramanters, filters... that I will like to host in the my tool. I need to build an AI based tool that can help me identify the reports that are in the the tool based on human based story that someone will write. Architecture:- User Story ↓ AI Extraction Layer ↓ Structured Matching Engine ↓ Vector / Semantic Search ↓ Report Definition Repository ↓ Explainable Response 20$ Per Milestone Milestone 1- Select Tools and Technology Create Initial Data Model Help me setup the initial setup upload to Data Model. Milestone 2- Build User story extraction flow When users submit a user story, first transform it into structured requirements. This extracted object becomes the input for matching with the data model. Train me on this step and how to learn this matching. Milestone 3:- Matching logic Using a weighted score as an example below Total Match Score = 35% Semantic intent similarity + 30% Column match + 15% Business object / data source match + 10% Filter / prompt match + 10% Functional area / project context match -- Help me manipulate and learn on how the above works. Milestone 4:- Build Match percentage bands Use simple business-friendly bands based on the score above to recommend likeliness of the report in repository or how much % it matches with the query details. -- Help me learn the process and how I can change it. Milestone 5- Sample output to end users Result: Strong match found — 82% Best matching report: Customer Invoice Aging Detail Why it matched: - Functional area matched: FIN - Business object matched: Customer Invoice - 6 of 8 requested columns were found - Existing report includes supplier, company, invoice date, due date, amount, and payment status - Existing report does not include cost center hierarchy or payment terms Recommendation :Reuse with modification. -- ****-- Help me learn how it works and how to make changes. Overall Flow:- 1. Reporting lead uploads report definitions from multiple projects 2. Platform parses and standardizes definitions 3. AI creates embeddings and searchable metadata 4. Business user enters a user story 5. Platform extracts intent, columns, filters, and business objects 6. Matching engine compares against existing definitions 7. Platform returns: - Existing report candidates - Match % - Columns found - Missing columns - Reuse / modify / build-new recommendation 8. Functional lead confirms decision Milestone 6 Once all Milestones 5 are completed successfully(Total 100$ with 20$ a Milestone) a one time bonus of 50$ to help create small documentation on all the milestones and also helping install and use the tool on a brand new Virtual Machine.