CS491 10/24

Introduction

  • Participants greet each other and confirm audio setup for the meeting.

  • Mention of a new audio setup without video presence.

Personal Updates

  • Speaker mentions personal life: son is a freshman and discusses the challenges of studying, particularly in Calculus.

  • Reference to studying habits and the importance of avoiding last-minute studying for exams.

Meeting Logistics

  • Offering to send meeting overview notes and attach PDF files.

  • Challenges with downloading documents noted due to government restrictions on the computer used.

  • Alternative proposed: convert documents to Google Docs for easier access.

  • Participants discuss whether screen sharing can work if document links are inaccessible.

Discussion on OCR Pipeline (Optical Character Recognition)

  • Meeting Overview Notes:

    • Participants discussed the discovery that ChatGPT includes built-in OCR capabilities, alleviating the need for them to create their own from scratch.

    • Advantages of Using ChatGPT:

    • Faster development pace

    • Improved accuracy in data extraction and preprocessing

    • Inquiry into whether others were surprised by ChatGPT’s capabilities: All agreed it was unexpected.

    • Importance of adapting to new technologies instead of re-creating solutions.

Data Processing Flow Diagram

  • Diagram Explanation:

    • Created by Jason to illustrate the flow from raw input data to structured JSON output.

    • Process Overview:

    • Raw data source

    • Orchestration Controller: Coordinates subsequent operations.

    • Preconditioning Service: Cleans data for processing.

    • Input Router: Determines processing type (text, image, multimedia).

    • LLM Gateway: Processes information.

    • Post-processing Service: Final adjustments and validations.

    • Output Validator: Ensures output quality and manages feedback loops back to the Orchestration Controller.

    • Text validation is crucial before outputting into JSON or other structured formats.

Feedback on Data Processing Strategy

  • Members acknowledged the importance of differentiating processing for structured vs. unstructured data.

    • Dave suggested separating the text processing chain for unstructured data.

  • Use of modular design is encouraged; flexibility in tools may simplify the development process.

Orchestration Service Design Overview

  • Diagrams and design documents for the orchestration service are discussed.

  • Technology Stack Overview:

    • N8N for workflow automation.

    • FastAPI for backend communication/API management.

    • Langchain with ChatGPT for AI processing and structuring input data.

  • Proposed workflow from N8N through processing and back to N8N for results.

  • Suggestions for future upgrades to scalable systems involve potential shifts to Flows and Llama Index.

Collaboration and Partnerships

  • Acknowledgment of potential licensing needs for specific tools (Langchain).

  • Discussion on evaluating the necessity for single or multiple user licenses for cloud environment deployment.

  • Participant inquiries about open-source alternatives or easier implementations.

Testing and Results of AI Models

  • Testing performed with ChatGPT and Claude.

  • Performance discrepancies noted with accuracy in recognizing dimensions on floor plans.

    • Floor Plan One showed better results than Floor Plan Two with ChatGPT; Claude performed inconsistently.

  • Conclusions drawn on using both AI models effectively maximizes accuracy in data extraction through joint utilization.

Project Management and Scheduling

  • Project Schedule Overview:

    • Deliverables are integrated for tracking progress and monitoring project advances.

    • Regular updates on schedule adjustments or areas needing more support.

    • Anticipation of future collaboration to ensure continued progression of project goals.

Next Steps and Closing Remarks

  • Participants encouraged to utilize ongoing adjustments for successful outcomes.

  • Emphasis on empowering team members for autonomous execution of tasks within tech infrastructures.

  • Request for a draft schedule for the next semester and acknowledgment of the current project status as ahead of schedule.

  • Encouragement to send any relevant updates on needs for data or tooling to facilitate project workflow.

  • Concluding with affirmations of teamwork and anticipation of the next meeting.