Skip to content

Examples

Hands-on Jupyter notebooks covering every major feature of Medha. Each notebook is self-contained and can be run locally or opened directly on GitHub.

git clone https://github.com/ArchAI-Labs/medha.git
cd medha
pip install "medha-archai[all]"
jupyter notebook demo/

Query Languages

Notebook Description
01 — Text-to-SQL (SQLite) End-to-end Text-to-SQL with a local SQLite database
02 — Text-to-Cypher (Neo4j) Graph query caching for Neo4j Cypher queries
03 — Text-to-Query (MongoDB MQL) MongoDB query caching with MQL
10 — Text-to-SQL (DuckDB) Analytical SQL caching with DuckDB

Embedders

Notebook Description
04 — Custom Embedder Implement BaseEmbedder for any embedding provider
07 — Cloud Embedders OpenAI, Cohere, and Gemini embedding adapters
05 — NER: spaCy vs GLiNER Parameter extraction strategies for template matching

Vector Backends

Notebook Description
11 — InMemory Backend Zero-dependency in-process caching
24 — Qdrant Backend Production-grade HNSW with Qdrant (memory / Docker / Cloud)
12 — pgvector Backend PostgreSQL-native vector search
15 — VectorChord Backend High-throughput PostgreSQL + VectorChord
14 — Elasticsearch Backend Full-text + vector search with Elasticsearch 8.x
16 — Chroma Backend Lightweight local vector store with ChromaDB
17 — Weaviate Backend Knowledge graph + vector search with Weaviate
18 — Redis Vector Backend Sub-millisecond semantic caching with Redis Stack
19 — Azure AI Search Backend Managed cloud vector search on Azure
20 — LanceDB Backend Embedded zero-infrastructure vector search

Features & Patterns

Notebook Description
06 — Production Patterns Logging, retries, health checks, and deployment tips
08 — Fuzzy Matching Tier 4 Levenshtein fallback for typos and variants
09 — Multi-Tenant Caching Separate collections per tenant
13 — Framework Integrations LangChain, LlamaIndex, and Haystack integration
21 — Cache Lifecycle TTL, expiry, and invalidation strategies
22 — Observability Stats, latency percentiles, and logging
23 — Batch Operations Bulk ingestion, export to DataFrame, and dedup
25 — Feedback Loop Recording correct/incorrect signals and auto-invalidation
26 — CLI All medha CLI commands against an in-memory backend