2025 · Team
QueryCraft
Describe what you need in plain English — get back a working SQL or NoSQL query.
Built with two other people (Syed Mohammed Sultan, Rifaque Ahmed Akrami — credited as authors in the real README) as a chat-based assistant that turns natural language into optimized database queries: it detects the intended query language, picks an LLM suited to the query's complexity, executes it against the target database, and keeps conversation context across follow-ups via automatic history summarization. Split across three repos — an umbrella repo plus separate frontend and backend — linked below.
Features
- Multi-LLM routing — Gemini, OpenRouter models (Mistral/Qwen/Llama/DeepSeek), and local Ollama, chosen by query complexity
- Multi-database execution — SQL (SQLite/PostgreSQL/MySQL), MongoDB, Neo4j, Cassandra, DynamoDB, Elasticsearch, GraphQL
- Smart query-language detection from natural language, with explicit-mention overrides ("use Cypher", "write MongoDB")
- JWT auth with bcrypt password hashing, rate limiting, and Helmet/XSS/CORS security headers
- File upload support — query directly against an uploaded CSV
- Conversation memory — automatic summarization of chat history to keep long sessions within LLM context limits
- Chat-based UI with persistent history, dark/light theme, and real-time typing indicators
- Response parsing that extracts executable queries from LLM output, including from code-fenced responses
Tech stack
Frontend
- Next.js 15
- React 19
- TypeScript
- Tailwind CSS
- Radix UI
- Framer Motion
- Three.js
Backend
- Node.js
- Express.js
- JWT auth
- Helmet
- express-rate-limit
Data
- MongoDB (Mongoose)
- PostgreSQL
- MySQL
- SQLite
- Neo4j
AI / LLM
- Google GenAI (Gemini)
- OpenRouter
- local Ollama