Assignment: INSPIREHEP Search & Summarization Web App

Setup Application

Backend

Prerequisites: Docker, Docker-compose

1. Rename ".env.example" file to ".env" and adjust
2. Setup
   • Run the django app

       make up 
       make bootstrap 
       # visit localhost:8000
       # user: admin
       # password: admin
   

The backend app should run on http://localhost:8000/ To access admin view: go to http://localhost:8000/admin with user and *password *

Celery Data Harvester

1. Run the backend
2. In another terminal copy the celery docker-container id by running

docker ps

4. Run celery worker

The harvester will fill the db with the api results in the interval set in CELERY_SCHEDULE_MINUTES in .env file (1 day default). Recommendation to put to 1 minute for test purposes! docker exec -it f <celery-docker-container-id> celery -A sis_exercise worker --loglevel=info
The command should return something like "celery@00769f42c709 ready." Depending on system sudo might be needed fro this command

5. Run celery beat docker exec -it <celery-docker-container-id> celery -A sis_exercise beat --loglevel=info The command should return something like "beat: Starting... Scheduler: Sending due task scheduler (api.tasks.harvest_literature)". Depending on system sudo might be needed fro this command

Frontend

Prerequisites: developed on node version v22.9.0

1. Go to frontend folder and rename frontend/.env.example to frontend/.env and adjust REACT_APP_BACKEND_URL if needed (default as above)
2. In frontend folder and run npm install npm start

The frontend app should run on http://localhost:3000/

---

Overview solution: Searching App

The solution provides a simple web application that allows users to search for high-energy physics papers using the INSPIREHEP REST API and receive a summary generated by the OpenAI API and the list of results. The application consists of a Django backend and a React frontend.

• The celery harvester harvests everyday the INSPIREHEP REST API, gets the first 40 papers and saves them
• The frontend UI provides an accessible search bar for the search queries and displays the OpenAI generated summary and the list of results
• The backend provides the endpoint for the search query, extracts titles and abstracts from the top search results, summizes the information using the OpenAI API or a mock function and returns it with the search results

Assumptions or shortcuts

• Publication date is not returned by the api for tested results but publication year. For the mocking of date assumed, the publication date is "01/01/year"
• Usage of ant.design for quick prototyping of the frontend
• Unit tests are not covering the use cases and are implemented as example
• For setup of frontend the simplest setup with react-create-app and js is taken. For real-world applications recommended setup with ts.
• cors policies allow all: would need to be changed for production

Given task: Create a task description for the following request:

• The product owner wants to have metrics about the OPENAI API response time and the most common user queries.
  • Create an issue for this task that you will give to your team to solve, please be specific. in the implementation.
  • OPTIONAL: Implement your suggestion.

Task Description: Metrics for OPENAI API Response Time and User Queries

• Task Title: Implement Metrics for OPENAI API Performance and User Interaction

• User Story: As a product owner, I want to gather metrics about the OPENAI API response time and the most common user queries displayed in the Django admin view so that we can analyze it's performance.

• Acceptance Criteria:

  • Response Time Measurement: Implement functionality to track the response time for each API call to the OPENAI API. Log response times in milliseconds for each request.

• User Queries Tracking: Capture and store all user queries and their counts sent to the OPENAI API. Ensure queries are logged with timestamps for tracking trends over time. Data Aggregation:

• Reporting Dashboard: Create a user-friendly dashboard in django admin panel that displays:

  • Logs with real-time response time metrics on each query (Openai api metrics)
  • A list of the user queries along with their frequency, average, max, min response time

• Error Handling: Implement robust error handling to log and report failed API requests separately.

• Documentation: Provide clear documentation on: How metrics are collected and reported. Instructions for accessing and interpreting the dashboard.

• Definition of Done:

  • Code is implemented, tested, and reviewed.
  • The metrics are being logged and can be accessed in the Django admin view via http://localhost:8000/admin/api/openaiapimetrics/ (logs with timestamps and response times per query) http://localhost:8000/admin/api/openaiapistatistics/ (statistics per query with count,avg, min, max response time)
  • The admin dashboard displays accurate metrics.
  • Documentation is updated and accessible.
  • Metrics are being logged and can be accessed by the product owner.

• Sprint considerations: Estimate story points for this task and prioritize it in the upcoming sprint. After implementation, gather feedback from the product owner to refine metrics further.

Overview solution: Metrics for OPENAI API

1. Run backend with instructions above
2. Access metrics http://localhost:8000/admin/api/openaiapimetrics/ (logs with timestamps and response times per query) http://localhost:8000/admin/api/openaiapistatistics/ (statistics per query with count,avg, min, max response time)

Constraints, assumptions or shortcuts

• All queries that return 0 results from Elasticsearch are not sent to openai and not displayed in the statistics