Rebuild EF Core pgvector hybrid search sample

[balajibashyam]

Summary

Replaces the existing minimal RRF-only Hybrid Search companion with the verified
.NET 10 implementation for
Hybrid Search in .NET with EF Core 10 and pgvector: Fuse Keyword + Vector Retrieval.

The existing GitHub path is preserved:

dotnet-ai/hybrid-search-ef-core-pgvector/

What's included

• Rebuilt sample

  • ASP.NET Core Minimal API
  • EF Core 10
  • PostgreSQL full-text search
  • generated tsvector
  • GIN index
  • pgvector vector(768)
  • HNSW vector_cosine_ops
  • Microsoft.Extensions.AI.IEmbeddingGenerator
  • Ollama nomic-embed-text default path
  • document ingestion/chunking
  • normalized + transactional document replacement
  • keyword, vector, and hybrid modes
  • Reciprocal Rank Fusion
  • 13 deterministic tests
  • Docker Compose
  • local end-to-end verification script

• Verification

  • .NET SDK 10.0.401
  • net10.0
  • Release build passed with 0 warnings / 0 errors
  • deterministic tests passed: 13/13
  • local PostgreSQL + pgvector live verification completed
  • local Ollama nomic-embed-text live verification completed
  • HTTP document ingestion verified
  • normalized document replacement verified
  • keyword retrieval verified
  • vector cosine retrieval verified
  • hybrid RRF retrieval verified
  • GIN index verified
  • HNSW vector_cosine_ops verified
  • stored vector dimension verified as 768

• Root README

  • existing Hybrid Search row updated
  • run/build/test commands updated
  • repository structure updated

• GitHub Actions

  • existing build-hybrid-search job updated for HybridSearch.slnx
  • deterministic tests added to CI
  • no Docker, Ollama, cloud credential, or live provider required

Package baseline

• Microsoft.EntityFrameworkCore 10.0.12
• Microsoft.EntityFrameworkCore.Design 10.0.12
• Microsoft.EntityFrameworkCore.Relational 10.0.12
• Npgsql.EntityFrameworkCore.PostgreSQL 10.0.3
• Pgvector.EntityFrameworkCore 0.3.0
• Microsoft.Extensions.AI 10.10.0
• Microsoft.Extensions.AI.OpenAI 10.10.1
• Azure.AI.OpenAI 2.1.0
• OllamaSharp 5.5.0

Evidence boundary

The local live publication path used:

• PostgreSQL/pgvector
• Ollama
• nomic-embed-text
• 768-dimensional vectors

OpenAI and Azure OpenAI adapter examples are compile/construction verified only.
No live cloud embedding request is claimed.

Deliberately omitted

• /api/ask
• chat/answer generation
• RAG prompting
• reranking
• authentication
• multi-tenancy
• dedicated vector databases
• automatic provider failover
• cloud live integration tests
• deployment/IaC

Database boundary

EnsureCreatedAsync is retained for reproducible clean tutorial databases.
Production applications should use deliberate schema migration/deployment practices.

Independence

This is an independent educational sample and is not an official Microsoft,
PostgreSQL, pgvector, OpenAI, Azure, or Ollama sample or specification.