Warlock AI

Langchain based agentic AI solution to support an engineering student's everyday university life.

Reason behind project

• To test / showcase Langchain / Python based approach in solving some LLM scenarios.

Features

1. General QA
   • LLM based chatbot functionality
   • Agentic support for coding (building, running, checking output) related queries via Python REPL
   • Agentic support for solving code error (bug, warnings, errors) related queries via StackOverFlow
   • Agentic support for queries with search engine requirement via DuckDuckGo
   • Blocking inappropriate (politics, terrorism, pornography) queries
2. Document Ingestion
   • Upload PDF or HTML documents into specific subject and topic to serve as context for future queries
   • User specific document handling
3. Document QA
   • LLM based chatbot functionality with context support based on subject and topic
   • User specific context handling

Arhitecture

The application is following a 3 service structure:

• warlock-ai-service -> Only AI related functions ("this repository")
• warlock-backend-service -> Classical backend related functions (looking for contributor due to time shortage)
• warlock-frontend-service -> Anything that is frontend (looking for contributor due to time shortage)

Current state

• This repository's scope is only to deliver the warlock-ai-service
• Functionality that should be implemented in warlock-backend-service is currently mocked by hardcoded values
• warlock-frontend-service is being mocked via CURL terminal calls
• Therefore, some features may be either missing (for example: chat history) or incomplete (like: validation, authentication, etc.)

Setup

Prerequisites

• Create a venv
• Run pip install -r requirements.txt

To develop warlock-backend-service / warlock-frontend-service

• docker compose -f compose/docker-compose-fs.yml up -d

To develop warlock-ai-service

• docker compose -f compose/docker-compose-ds.yml up -d

Examples

1. General QA

   1. Block political, terrorism and pornographic queries
      • curl -X POST http://127.0.0.1:8123/api/python/v1/general-qa -H "accept: application/json" -H "Content-Type: application/json" -d '{"query": "Was Bush a great president?", "warlock_api_key": "EMPTY"}' -N
   2. Trigger Coding Agent
      • curl -X POST http://127.0.0.1:8123/api/python/v1/general-qa -H "accept: application/json" -H "Content-Type: application/json" -d '{"query": "What is the length of the final array, when you create an empty array and append numbers [1, 2, 3] to it?", "warlock_api_key": "EMPTY"}' -N
   3. Trigger StackOverFlow Agent
      • curl -X POST http://127.0.0.1:8123/api/python/v1/general-qa -H "accept: application/json" -H "Content-Type: application/json" -d '{"query": "How to solve error: TypeError: can only concatenate str (not int) to str?", "warlock_api_key": "EMPTY"}' -N
   4. Trigger DuckDuckGo Agent
      • curl -X POST http://127.0.0.1:8123/api/python/v1/general-qa -H "accept: application/json" -H "Content-Type: application/json" -d '{"query": "Which date is Independence Day on?", "warlock_api_key": "EMPTY"}' -N
   5. Trigger vanilla LLM
      • curl -X POST http://127.0.0.1:8123/api/python/v1/general-qa -H "accept: application/json" -H "Content-Type: application/json" -d '{"query": "What is my name?", "warlock_api_key": "EMPTY"}' -N

2. Document Ingestion

   1. PDF
      • curl -X POST http://127.0.0.1:8123/api/python/v1/document-ingestion -H "accept: application/json" -H "Content-Type: application/json" -d '{"user_id": 111, "subject_id": 1, "topic_id": 11, "document_path": "/Users/mac.username/Documents/document_ingestion_test.pdf", "document_id": 101, "document_type": "PDF"}' &
   2. HTML
      • curl -X POST http://127.0.0.1:8123/api/python/v1/document-ingestion -H "accept: application/json" -H "Content-Type: application/json" -d '{"user_id": 111, "subject_id": 1, "topic_id": 11, "document_path": "http://car-database-bucket.s3-website-eu-west-1.amazonaws.com", "document_id": 102, "document_type": "HTML"}' &

3. Document QA

   1. Use ingested document to get answer
      • curl -X POST http://127.0.0.1:8123/api/python/v1/document-qa -H "accept: application/json" -H "Content-Type: application/json" -d '{"query": "What is the color of the car with license plate number UET-218?", "warlock_api_key": "EMPTY", "subject_id": 1, "topic_id": 11}' -N