ZigSmartSearch: Fast Semantic Search in Zig

[hybrid kestrel]

🔍 Explore concept-based search for JSON datasets with these key advantages:

• 📖 Dictionary-Based Expansion: Understands meaning via a custom JSON dictionary of weighted related terms.
• 🧠 Lemmatized Matching: Captures word variations using base-form mapping for better relevance.
• 🎯 Precision Scoring: Rewards exact/phrase matches, penalizes irrelevant terms, and normalizes for fairness.
• ⚡ Fast Performance: Powered by Zig for quick, memory-efficient searching.
• 🔌 No Dependencies: Pure Zig implementation, keeping it lightweight and self-contained.

⭐ Please Star the repo to support the project: https://github.com/RohanVashisht1234/ZigSmartSearch

[marble forum]

Sound a lot like what a search engine would do ? I mean if it was a great search engine not done by evil corp

[lime burrow]

1. Too much vibe-coded.
2. SOTA approach is to use neural network/autoencoders/BERT to get text embedding and use similarity metrics like dot product between query and the database.

[hybrid kestrel]

Yes, It is similar to a search engine by ranking results by relevance. But this is a lightweight, no dependency tool written entirely in Zig. It searches raw JSON using custom dictionaries and lemmatization, with no external services or heavy infrastructure.

[hybrid kestrel]

(1) I used AI tools for basic formatting and a few helper functions, but Zig is not well supported by LLMs yet at a level to "vibe code", so the core logic i.e dictionary design, lemmatization, matching, was fully implemented by me manually.
(2) Deep learning models like BERT would break the "pure Zig, zero dependencies" goal.

[lime burrow]

What if training of BERT performed using trainer that were implemented in Zig, does that count? ----- Think of it as Zig depends on Git, HTTP, TLS 1.3, etc. BERT is just float numbers with math ops, and inference based on Zig code.

Zig should be able to implement universal function approximators, otherwise you are dealing with your delusions

About first point - try to Google for "aistudio Google"