[Feature] Meaningful Code Summaries using Gen AI

[parinaB]

Problem Statement

The summary field in the /explanation/ endpoint is generated from a hardcoded template and looks like this:

"A short Python snippet (5 lines) that performs a focused task."

The engine never actually reads the code — it just fills in line count into a fixed string. This gives developers zero useful information about what their code actually does, which defeats the entire purpose of the explanation endpoint.

Proposed Solution

Implement a three-tier summarization system with graceful fallback:
Tier 1 — Gen AI (Best quality, requires API key)
When LLM_ENABLED=true, route the summary through the existing ai_provider.py abstraction layer with a focused prompt

Alternatives Considered

Offline NLP model (Good quality, no API key)

When NLP_ENABLED=true, use a BART or T5 model fine-tuned on CodeSearchNet for offline code-to-text summarization:

[Kalpnaa]

Hello @imDarshanGK

I would like to work on this issue under GSSoC'26.

I reviewed the problem statement and agree that the current /explanation/ summary generation is too generic since it relies on a hardcoded template instead of analyzing the actual code. This reduces the usefulness of the endpoint significantly.

Approach

I plan to implement a three-tier summarization system with graceful fallback:

Tier 1 — Gen AI Summarization**

• If LLM_ENABLED=true, route summary generation through the existing ai_provider.py abstraction layer.
• Use a focused prompt to generate concise and meaningful code summaries.

Tier 2 — Offline NLP Summarization**

• If NLP_ENABLED=true, integrate an offline summarization model (BART/T5 or lightweight transformer trained for code summarization).
• This ensures quality summaries even without external API access.

Tier 3 — Rule-Based Fallback**

• If AI/NLP models are unavailable, implement static code analysis using AST/parsing.

• Extract key signals such as:

  • function/class names
  • imports
  • major operations
  • line count

• Generate a smarter summary instead of the current hardcoded string.

I can start working on this immediately and will ensure clean, well-documented implementation.

Kindly assign this issue to me.

Thank you!

[parinaB]

@imDarshanGK any updates??

[imDarshanGK]

@parinaB Thank you for your interest. This issue is available to work on.

Please keep the implementation lightweight and focused on improving the summary generation using the existing architecture. Avoid adding heavy external NLP/transformer model dependencies unless necessary.

Assigned.