Insight

Question Answering

• bloop - fast natural language code search written in Rust
• Sourcegraph Cody
• Repo Chat by Mckay Wrigley

Doc Gen

• GptDoc
• Craft vs Cruft episode - Experimenting with GPT3.5 generating Doxygen comments for the Duke Nukem 3D code

Architecture Mining

• FlowGPT - Generate flow charts from code

SQL Interfaces

• SQL Chat - Chat-based SQL Client and Editor.
• PG GPT - Postgres extension that generate SQL queries from natural language, and converts query plans to natural language.
• LangChain SQLDatabaseChain - let an LLM agent interface with a SQL database.

Semantic Code Search

Semantic search attempts to consider the contextual meaning of the terms, as opposed to traditional lexical search which looks for literal matches of the query words and simple variants. In the context of language models, this is usually implemented with embeddings, but this has not been widely adopted for source code.

• GitHub research blog
• CodeSearchNet - DataSet and Benchmark for training semantic code search models
• sem CLI tool
• Source Graph - A leader in corporate code search. Here they use 'semantic' to mean 'language-aware', not using embeddings.