query-routing

Low latency, High Accuracy, Custom Query routers for Humans and Agents. Built by Prithivi Da

Sub-millisecond query routing & cost-aware execution for GraphRAG architectures. Adaptive-AtomRAG is a highly optimized conditional routing gateway designed to resolve the latency and cost bottlenecks of heavy Atom-Entity Graph (AEG) RAG systems.

LLM reasoning: use draft agreement to route queries and allocate adaptive thinking budgets without training.

DuckDB extension for system monitoring & intelligent SQL routing. 25+ functions for CPU, memory, disk, network, processes. Conditionally Route queries to PostgreSQL, MySQL, Snowflake, BigQuery with automatic dialect translation via SQLGlot. Pure Rust, cross-platform.

An intelligent, stateful Customer Support Agent built with LangGraph and LangChain. It categorizes incoming queries, analyses their sentiment, and either routes them to the right specialist handler or escalates negative-sentiment queries to a human agent.

Reference multi-tenant knowledge-base system using LlamaIndex, RAG, vector search, SQL generation, and query routing.

Reference architecture for legal contract intelligence, dual runtime (Azure / local docker)

This project demonstrates the integration of multiple AI agents using Retrieval-Augmented Generation (RAG) and WikiSearch functionalities, along with Cassandra for data storage and retrieval.

Routes incoming prompts to the cheapest model tier that can handle them correctly, reducing inference cost while maintaining quality

Multilingual decision engine for intents you describe in plain words, no training examples. It picks the matching intent, with a margin that flags doubtful answers; R2Intent v1 also answers yes or no (beta), with probabilities. About 26 ms per message on a GPU (compiled). Python, HTTP or MCP, self-hosted. Library Apache-2.0; models CC BY-NC 4.0.

Adaptive RAG pipeline with a lightweight DistilBERT query router for dynamic, latency-optimized retrieval strategies.

QLoRA Gemma-3-1B router for adaptive RAG strategy selection (DIRECT / PARALLEL / SEQUENTIAL). M.Eng. thesis.

Routes queries to PostgreSQL, Redis, or pgvector by learning query intent instead of relying on hardcoded rules.

Database connection pooler and query router for the Servverse ecosystem — read/write splitting, slow query detection, multi-dialect support.

Top-k retrieval cannot count, detect absence, or compare magnitudes. A query router that dispatches to engines that can: 1.000 vs 0.156 against a naive RAG baseline on 24 golden queries.

Adaptive RAG over 10 RAG research papers: a fine-tuned BERT router (96.7% accuracy, 13.8 ms/query) sends each question to a direct LLM answer, web search or corrective RAG. Benchmarked against a 4B LLM router (JEV), evaluated with RAGAS, served with FastAPI + Streamlit in Docker.

Agentic RAG toolkit over your own PDFs built with LlamaIndex: Query Router (summary vs. vector retrieval), Tool Calling with page-filtered Vector Search, a Reasoning Agent with memory, and a Multi-Document Agent with Tool Retrieval. OpenAI LLMs + Embeddings, RAGAS evaluation, Python library + CLI.

High-performance Hybrid RAG pipeline using LlamaIndex, MinerU, GLiNER, and LangExtract. Features advanced metadata enrichment, BM25-Vector fusion, and LaTeX-enabled synthesis for complex document analysis.

An intelligent LLM routing system optimizing the cost-quality trade-off for RAG pipelines. Uses multi-tier semantic complexity scoring and embedding models to dynamically route queries.

A domain-agnostic, zero-dependency adaptive query router for RAG systems. Optimizes retrieval strategies using multi-signal statistical analysis of pilot scores.