Vajra: step-by-step programming with natural language
summary
Vajra converts short natural-language commands into runnable Python by extracting semantic roles with off‑the‑shelf parsers, generating API-aware candidate statements constrained by current code state, and ranking them with feature vectors built from string and embedding distances and an SVM ranker. It follows a step-by-step single-statement generation model rather than end-to-end synthesis, is nearly training‑free and aimed at custom APIs, and was tested with 12 users (6 programmers, 6 non‑programmers) on 6‑ and 11‑line tasks with successful completion and a SUS 85.4 usability score.
AI-written summary · about
chapters (6)
- 0:00Start
- 1:34Programming with Natural Language
- 4:12Perform Semantic Parsing
- 4:52Semantic Parsing
- 6:45Candidate Ranking
- 13:16Is Complex Natural Language Understanding Necessary
related talks
24mNatural Language Coding Revolutionising Programming with GenAI
25mNo Code, No Problem: Building Powerful AI Agents with Just Your Words
46mThe Era of AAP: Ai Augmented Programming using Modern Java
15mNatural Language-Guided Programming
41m"Natural Interfaces" and "Natural Language Programming"
12mNavigating the World of AI Development Without Going Insane
38mThe Era of AAP: Ai Augmented Programming using only Java
10mApp-Based Task Shortcuts for Virtual Assistants