This week, we'll focus on documenting the code and preparing for the final presentation.

The ninth week essentially consisted of integrating a new functionality to upload relevant PDF files into a vector store for improved responses. 

This allows the system to prioritize information from stored PDFs, falling back on internet searches only if the information retrieved from the vector store is irrelevant. The upload feature is restricted to certain admin users at Reach for Change.

Moreover, the option of copying the response or downloading a PDF of the response from the interview scripter and survey generator was also added.

In the eighth week, we made significant aesthetic improvements and integrated new pages for features like the interview scripter and survey generation. 

Enhanced the overall look and feel of the site, adjusted back end processes, and tried to streamline the deployment using CI/CD and FastAPI for cleaner, easy-to-use code.

This week, we enhanced backend functionalities and introduced new ones. We integrated Brave for search to better prioritize authoritative webpages and refined the prompt for the interview scripter. 

Additionally, we resolved citation errors and worked on effectively integrating the survey generator into the platform.

This week, Robert focused on refining citation accuracy in LLM-generated responses, ensuring that all sources were correctly tagged with citation numbers.

Additionally, he developed a survey generator by crafting a custom prompt that outputs questions in JSON format.

In the fifth week of development, our team has made significant progress in enhancing the web application:

1. User Authentication: Implemented a log-in system to improve security and control access.
2. UI Upgrade: Migrated the entire user interface to React, resulting in a smoother user experience.
3. New Specialized Pages: Added several purpose-specific pages to expand functionality:
   • Home
   • About
   • Interview Scripter
   • Research Assistant
   • Root Cause Analysis
   • Consequence Tracking

These improvements have not only strengthened the app's security but also enhanced its usability and feature set, providing a more comprehensive and user-friendly experience.

This fourth week basically entails improving both the backend and the frontend of our product.

We started making sophisticated modifications to our backend, including accessing the World Bank API for reliable information about socioeconomic issues and using Tavily's API to retrieve trustworthy URL links.

We also shifted our focus to making our webpage not just functional but also visually appealing and user-friendly.  In this way, we would like to improve the design further, by using React as a better front-end framework and adding new features to enhance the user experience.

In the third week, we agreed on the final architecture and started with the first prototype of the webpage.

• Frontend with Amplify: We opted for AWS Amplify to build our frontend.  It allowed us to create a user-friendly interface and effectively manage our web resources.
• Backend with Lambda Functions: Our backend relied on AWS Lambda functions.  We used them for tasks like running the backend and handling REST API routes.
• Storing Data in S3 Buckets: To store data, we leveraged Amazon S3 buckets. These reliable storage containers held our scrapped PDFs.
• API Gateway for Communication: Our API Gateway acted as the bridge between our front end and back end. It facilitated communication, ensuring interactions across different parts of our application.
• Scraping Reliable Sources: Our process involved scraping information from trustworthy internet sources. Tools like RAG and LLMs helped us extract relevant data efficiently.

After implementing the whole pipeline, we managed to upload our first product and go live. We got the first look at our webpage with all the minimum requirements in place.  It was an exciting transition from planning to execution, and we couldn’t be happier with the results we achieved together!