Code Comprehensor: Hierarchical Documentation and Cross-Modal Retrieval for Large Repositories
We propose Code Comprehensor, a hybrid retrieval-reasoning framework combining (i) a code dependency callgraph, (ii) CodeScribe, which uses code-aware LLMs to generate hierarchical documentation across repository call graphs, and (iii) CodeContrast, a CLIP-style dual-encoder model trained on CodeScribe output to learn aligned text-code embeddings.
ProofPaper ↗
Key points
- Understanding large, complex, and rapidly evolving code bases remains a core challenge for software engineering at scale, hindering onboarding, maintenance, refactoring, and impact analysis.
- CodeContrast enables fast, semantically precise retrieval of relevant code-context neighborhoods, which serve as grounded evidence for an MCP-based LLM orchestrator answering domain-specific queries.
- Evaluation on OpenCV and a proprietary 5M-LOC repository shows that CodeScribe-derived supervision significantly improves cross-modal retrieval in text-to-code accuracy over baseline approaches.
- Richer textual supervision (hierarchical documentation, parent/sibling context, and LLM-generated synthetic questions) consistently enhances performance across both repositories.
Sources (1)
- [1]Code Comprehensor: Hierarchical Documentation and Cross-Modal Retrieval for Large RepositoriesarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 04:45 PM
We propose Code Comprehensor, a hybrid retrieval-reasoning framework combining (i) a code dependency callgraph, (ii) CodeScribe, which uses code-aware LLMs to generate hierarchical documentation across repository call graphs, and (iii) CodeContrast, a CLIP-style dual-encoder model trained on CodeScribe output to learn aligned text-code embeddings.
Understanding large, complex, and rapidly evolving code bases remains a core challenge for software engineering at scale, hindering onboarding, maintenance, refactoring, and impact analysis.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026A Mechanistic Study of Automated Failure Attribution in LLM-based Agentic Systems
- Oct 20, 2026Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets
- Oct 20, 2026Securing AI agent identity with X.509 PKI on AWS
- Oct 20, 2026[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
- Oct 19, 2026Always On - A collaboration between two LLMs
- Oct 19, 2026Introducing Clairic, a new System One class of extraction models