Subject
11 entries
Code Generation
Bookmarks
Vibe-DSPy: Natural Language to DSPy Signatures
Vibe-DSPy converts natural language descriptions into DSPy signatures automatically, with iterative refinement from feedback. Takes the vibe-coding approach to generating structured LLM pipeline components.
monitors4codegen — Monitor-Guided Decoding
Monitor-Guided Decoding uses LSP (Language Server Protocol) static analysis as a 'monitor' during code LM generation to enforce semantic validity — identifiers must exist, types must match. NeurIPS 2023 paper from Microsoft with the multispy Python library for building LSP-backed code gen applications.
DSPy Self-Discover Framework
A DSPy + Groq-powered framework that takes natural language task descriptions and generates executable Python code to solve them — automated problem-solving through AI-driven reasoning and code generation. Now archived.
LLM Verified with Monte Carlo Tree Search
A research project using Monte Carlo Tree Search to guide and verify LLM code generation — MCTS explores the tree of possible completions and selects branches where generated code actually passes verification checks. An early example of search-augmented LLM reasoning.
backend-GPT: Natural Language Backend Generation
backend-GPT is an early experiment in using GPT to generate backend code from natural language descriptions — part of the wave of GPT-powered code generation tools that emerged before GitHub Copilot popularized the category. Represents the early exploration of LLMs as backend architects.
PAL: Program-aided Language Models
PAL prompts LLMs to generate Python programs as intermediate reasoning steps, then delegates actual computation to an interpreter, outperforming chain-of-thought on math and symbolic tasks. It neatly separates what LLMs are good at (translating problems to code) from what they're bad at (arithmetic).
OpenAI Codex TodoMVC Challenge
A walkthrough of using OpenAI Codex to build a TodoMVC app from natural language descriptions. Captures the pre-Copilot moment when Codex-powered code generation was the frontier of AI-assisted coding.
Language Models Can Teach Themselves to Program Better
Haluptzok, Bowers, and Kalai show that language models can generate their own programming problems and solutions, verify correctness with a Python interpreter, then fine-tune on the verified examples — more than doubling test accuracy. A clean demonstration of self-improvement via external verification.
FauxPilot: Open-Source GitHub Copilot Server
FauxPilot is an open-source, self-hosted alternative to GitHub Copilot that runs Salesforce's CodeGen models locally via Triton Inference Server. Built in response to privacy concerns about Copilot sending code to OpenAI's servers.
MutableAI: AI-Powered Code Generation
MutableAI is an early AI coding assistant that offered automatic code completion and refactoring — a 2022-era precursor to what Copilot and Cursor would become. Interesting artifact of the early AI coding tools moment.
jupyter-text2code: English to Python in Jupyter
jupyter-text2code is a proof-of-concept Jupyter extension that converts English queries into Python code — an early 2021 demonstration of natural language to code generation in notebooks, before Copilot and ChatGPT made this mainstream.