HL V1 Mediation App Final Outline

Overview

Overall, V1 aims to provide Hepworth with a low-cost version of the Elerion Mediation App that can easily change and adjust features and functionality as insights are gained from real-world application. This will be achieved by:

  1. Building the initial frontend website using WordPress.
  2. Using the Airtable CRM for admin user management, security, communication, and follow-ups, which will function as the main database.
  3. Using the Airtable Portal on the site connected to the Airtable CRM for user login, inputs, changes, and interactions before and after the mediation.
  4. Keeping complex development contained to:
    1. Extending livekitt meetings functionality.
    2. Adding a semantic search layer between TTS and AI Response.
    3. Fine-tuning the GPT model to provide mediator responses.
    4. Implementing draft agreement functionality.

The following outline breaks up all the work and time required into three categories:

  1. Frontend Website, Airtable CRM & Portal
    1. Estimated time to complete: 125 hours
  2. Meeting Functionality
    1. Estimated time to complete: 84 - 100 hours
  3. AI Fine Tuning
    1. Estimated time to complete: N/A - Artem completing planning

Please note that time is the maximum amount of time it should take to develop all listed features.

1. Frontend Website, Airtable CRM & Portal

Estimated time to complete: 125 hours

Summary of functionality to be created:

A front-facing website will advertise Elerion and its services, consisting of a home page, features page, pricing page, and sign-up page.

A secure portal will allow signed-up users to log in and access additional features, including the ability to add more users for the meeting, schedule a meeting time with the other primary user, add pre-mediation notes, obtain meeting access links, pay for mediation after it is completed, and view draft agreements. User management, follow-up, and the database will be easily controlled by Hepworth because it will be built around Airtable.

The following task list will achieve the below features:

  • Frontend website
  • Payments
  • Airtable CRM to manage users and database
  • Portal on frontend website connected to airtable for user management
  • User management, primary users that can add basic users
  • Add and update mediation information before it happens
  • Schedule mediation time between two primary users
  • Notifications on actions and upcoming meeting using existing Airtable CRM functionality

Detailed task list:

  • (20 hours) Setup Elerion CRM Airtable instance
    • Setup users table
      • ‣
      Users Table Outline
    • Setup meetings table
      • ‣
      Meeting Table outline
    • All follow-ups from any actions logged into interactions
      • Setup functionality for follow-up tasks to trigger X time from a date fields in Airtable (used for meeting notifications)
  • (20 hours) Setup front facing website to market Elerion: homepage and features
  • (5 hours) Setup sign up page that contains gravity forms and stripe add-on to process payments
    • Fields:
      • Party 1 Name, Email, Phone
      • Party 2 Name, Email, Phone
    • Setup submission submission Zapier listener:
      • Create new primary users in Airtable
      • Airtable triggers email and text message send-out notifying both parties to login
        • Second primary user will be prompted to pay once logged in
      • Person on sign up is redirected to portal login
  • (5 hours) Setup Airtable Portal on website
  • (15 hours) In Portal - Main landing page - Schedule meeting time functionality
    • Both users are presented gravity form to submit 3 preferred times
    • On submission from both users times are sent to GPT to generate 4 recommend times, 5th option that no times work
    • Recommended times stored in airtable, trigger notifications to users to login to portal to select all times that work.
    • If no matching times selected then repeat process
    • If matching time selected then stored as meeting time in airtable
    • When meeting time updated then /generate-meeting endpoint hit to store meeting links and passcodes in airtable
  • (10 hours) In Portal - Additional tab - Party Information
    • A main user will be able to add / edit mediation information before it starts
    • Use a gravity form that shows the submission data
  • (15 hours) In Portal - Additional tab - Add basic users functionality
    • A primary user will be able to add and delete Basic Users
    • Name, Email, Phone, Role
  • (10 hours) In Portal - Callout - Join Meeting
    • Time of meeting
    • Join meeting button
    • Meeting links are returned and stored in airtable and presented to user
    • Access meeting links
    • Generate new meeting links
  • (10 hours) Airtable CRM - Upcoming meeting notifications functionality
    • Notifications sent using airtable CRM at intervals before meeting time occurs, can be managed easy using follow-up tasks
    • Will need to build notifications from a certain time functionality
  • (15 hours) Airtable - Meeting ended, payment functionality
    • Airtable automatically update:
      • Meeting Duration
      • Meeting Transcripts
      • Meeting Audio
    • Update portal view to stripe payment form with amount required based on meeting duration and rate in airatble

2. Meeting Functionality

Estimated time to complete: min:119 hours max:145 hours (detailed+langchain)

Summary of functionality to be created:

Elerion-Backend project is based on 2 main structures. The first one is the LLM infrastructure that answers the questions and the other one is STT and TTS.

The main objectives of this study are as follows

  • Separation of LLM and STT/TTS structure
  • The front-end exchanges data with the LLM via API
  • Administrator interface that will run on AirTable

The following task list will achieve the below features:

  • By realizing the data exchange with LLM via API, our frontend and backend will be separated from each other.
  • Meetings will be password-protected, participants will be identified and the duration will be limited.
  • The ability to work with multiple LLMs at the same time will be gained.
  • Token limit-oriented working capability will be gained.
  • It will be able to grow horizontally through structures such as load balancer with API infrastructure.
  • When the mediation meeting is completed, all conversations will be uploaded to AirTable in PDF format.

The infrastructure will be made ready for future add-ons.

  • Summarization
  • Ability to use vector database
  • Access to various document sources
  • Ability to read data such as PDF, MarkDown, JSON, HTML
  • Ability to read data from RDBMS databases via SQL

Detailed task list (min:94 hours max:114 hours)

  • (5 hours) Removing ChatGPT from Elerion-Backend repository
  • (3 hour) Preparation of the development environment of the Elerion-AI code repository
  • (3 hours) Rest API, preparation of authorization (Basic AUTH) infrastructure
  • (3 hour) Preparation of AirTable infrastructure
  • (5 hours) /generate-meeting endpoint
  • (3 hours) /active-llm endpoint preparation
  • (10 hours) /completions preparation of the endpoint
  • (15 hours) Making Elerion-Backen work with Elerion-AI
  • (5 hours) Preparation of Meeting Time Control structure
  • (3 hours) Adding necessary components to the meeting enterance screen
  • (3 hours) Uploading meeting mp4 recording to S3
  • (5 hours) All conversations saved to RDBMS
  • (5 hours) All conversations uploaded to AirTable in PDF format
  • (5 hours) Sending information such as meeting duration and transcribe to AirTable
  • (5 hours) Ensuring data exchange of API with ChatGPT
  • (5 hours) Ensuring data exchange of API with Calude
  • (5 hours) Ensuring data exchange of API with ChatBase
  • (5 hours) Standardization of the result data to be returned to Elerion-Backend
  • (1 hour) Ensuring that the Elerion-AI project works as a docker service

LangChain task list (min:25 hours max:31 hours)

  • (10 hours) Learning the capabilities for LangChain integration
  • (5 hours) Using LangChain for the /completions endpoint
  • (5 hours) Assigning LangChain the task of writing the meeting content to RDBMS (not 100% sure)
  • (5 hours) ChatBase is not available in LangChain because it is not a real LLM. If LangChain structure is to be used, ChatBase integration should be done in addition to all integration.

3. AI Fine Tuning