ACM Summer School on Generative AI for
Text
Hackathon: Development of a PDF reading application using LLMs
Total Points: 100 Submission deadline: 23:59:59 Hrs, July 02, 2024
Hackathon Grading Instructions
Please refer to the following assignment instructions:
1. No queries will be answered after July 2, 2024, 11:00 IST.
2. We will follow the zero plagiarism policy, and any act of plagiarism will result in a zero for the
hackathons. Do not use any existing LLM-backed PDF question-answering pipeline, model, or
agent.
3. We will share a slot to showcase your solution to the Team. In addition, you have to submit the
video of the web application and the code of the working solution to GitHub and share the link
with us.
4. The GitHub repository should include a readme.md with instructions to setup the project. Also
include a requirements.txt with all libraries.
5. Please fill out the Google Form to submit your GitHub repository and video. Only one team
member needs to fill the form.
Problem Statement
Design a web application (you can use any technology) with a simple query interface that answers
queries grounded on the following two PDFs:
i. Details of Electoral Bonds submitted by SBI on 21st March 2024
(EB_Redemption_Details) [Bonds encashed by political parties]
ii. Details of Electoral Bonds submitted by SBI on 21st March 2024
(EB_Purchase_Details) [Bonds purchased by Individuals and Companies]
Technical Details:
i. The chat Interface should be similar to Gemini, chatGPT, etc.
ii. Connect it with existing LLMs via free APIs, such as Groq, GPT, Gemini, etc.
iii. You are free to choose any hosting platform. You can even host it on your laptop.
iv. The web application should be able to answer questions (user prompts) from the
PDF.
v. The response should only contain numbers or numbers with denominations.
vi. In addition to the chat interface, also submit a python script that can read
questions from a text file (questions.txt, each line contains a question) and write
answers to a file answers.txt.
Evaluation Guidelines:
i. Navigability and ease of use of web Interface. [10 points]
ii. Accuracy of answers on 10 test queries. Binary marking. [8*10 = 80 points]
iii. Innovation in pipeline integration, code, cleanliness of code, etc. [10 Points]
Sample Queries and Responses:
Q1: What is the total bond amount enchased by TELUGU DESAM PARTY on 12th April 2019?
R1: 3000000 or 30 Lakhs
Q2: What is the total bond amount purchased by CHOUDHARY GARMENTS on 12th April
2019?
R2: 2000000 or 20 Lakhs
Q3: What is the total number of bonds purchased by ACROPOLIS MAINTENANCE SERVICES
PRIVATE LIMITED on 12th April 2019?
R3: 7
Q4: What is the total amount received by AAM AADMI PARTY from DR. MANDEEP
SHARMA in the year 2022?
R4: 1000000 or 10 Lakhs.
