Skip to content

services.md

Latest commit

 

History

History
41 lines (28 loc) · 2.52 KB

services.md

File metadata and controls

41 lines (28 loc) · 2.52 KB
Raw

Problems Services

Because the Problems module is incredibly dense, its services are divided into two distinct categories: Standard Logic and AI Generation Engines.

File Location: api/src/services/problems/

1. Standard Logic Services

problem.service.ts

The core CRUD manager. It bridges the problem.repository with the API payload. Nothing fancy, just clean, validated data access.

problem-test.service.ts

The test case manager.

  • Security: When getTests is called by a regular user, this service actively filters out any test cases marked as type === "hidden", ensuring users cannot scrape the hidden tests used by the Judge module to evaluate their code.

problem-validator.service.ts

A strict syntax validation service. It uses acorn (a JS parser) to parse function_signature or class_signature JSON payloads and ensure that the provided syntax is valid JavaScript/TypeScript before saving it to the database.

2. AI Generation Engines

These services utilize LLMs to automate the heavy lifting of content creation.

ai-problem.service.ts

  • Engine: LLM via One API Gateway
  • Role: Takes a text prompt (e.g., "Create a Leetcode Medium about Two Pointers") and returns a fully structured JSON problem document.
  • Complexity: Highly complex system prompt engineering to enforce exact JSON schemas (title, description, difficulty, tags, function signature, boilerplate code in 7 languages).

testcase-generator.service.ts

  • Engine: LLM via One API Gateway
  • Role: Automatically generates 15-20 robust edge-case test validations for a given problem signature.
  • Execution: Because the AI only generates the inputs, this service actually calls the backend CompilerService to run the AI's inputs against the official solution to mathematically generate the expected_output!

testcase-canonical.ts

  • Engine: LLM via One API Gateway
  • Role: A parsing assistant. If an admin pastes raw text from a website like Codeforces, this service uses the LLM to intelligently extract the inputs and expected outputs and convert them into the strict [{ input, expected_output }] JSON array required by our database.

ai-addsolve.service.ts

  • Engine: AWS Bedrock (Anthropic Claude 3.5 Sonnet)
  • Role: The ultimate solver. It takes a newly generated problem and uses Claude Sonnet's superior reasoning capabilities to write the perfect canonical_solution (official solution) for it. It then updates the database directly.