1. Defined the project intention: organize rules into a consumable/learnable format and create a tool to easily validate the rules against advisor marketing collateral. The audience is for my wife with an eye towards positioning it for her advisors.
2. Convert rules: I converted her disorganized rules document into usable rule objects, from:
> (for all tax related references/ Tax disclosure)
> [FIRM NAME] does not render legal, accounting, or tax advice. Please consult your tax or legal advisors before taking any action that may have tax consequences.
into:
{
"title": "Tax Disclosures",
"slug": "tax-disclosures",
"original_text": "[FIRM NAME] does not render legal, accounting, or tax advice. Please consult your tax or legal advisors before taking any action that may have tax consequences.",
"summarized_text": "Consult tax/legal advisors before actions with tax consequences.",
"parent_rule": null
}
3. Educational portion of UI: I built and organized the rules into the UI and fought with an experimental vuetify plugin for far too long to display the rules in an easily consumable manner.
4. Rules validation tool (UI): Accepts an uploaded/dropped file and displays any rules that have been successfully or unsuccesssfully validated by the backend API.
5. Rule validation tool (API): Parses the incoming file into two different extraction pathways, either text extraction from a document or base64 extraction from an image. The extracted file info and the ruleset are passed to OpenAI api (using vision with an image) with a prompt looking for a list of all applicable rules that are either succcessful or unsuccessful, as well as a brief summary of the result.
6. Deployment: AWS Amplify & API Gateway/Lambda
That's about it. Currently waiting for feedback for any future iterations...
