Fix building path

[openminddev]

This pull request introduces a new embedding microservice based on FastAPI, optimized for NVIDIA Jetson (ARM64), and integrates it into the deployment and QA infrastructure. The changes automate the build and release of the ARM64 embedding service Docker image, add a new embedding service to Docker Compose, and provide a production-ready QAEngine that uses this service for fast, GPU-accelerated question answering.

Embedding Service Introduction and Integration

• Added a new FastAPI-based embedding microservice in embedding_server.py that exposes /embed, /embed_batch, and /health endpoints for real-time and batch text embedding using the intfloat/e5-small-v2 model on CUDA. The service returns base64-encoded float32 vectors and is optimized for low-latency inference on NVIDIA Jetson devices. (src/embedding/embedding_server.py)
• Created a dedicated Dockerfile for the embedding service (Dockerfile.embed), installing all dependencies, downloading the model at build time, and setting up the service to run on port 8100. (docker/Dockerfile.embed)

Deployment Automation and Infrastructure

• Updated the GitHub Actions release workflow to build and push the ARM64 embedding service Docker image, and to create/push its manifest for multi-architecture support. (.github/workflows/release.yml) [1] [2] [3]
• Added the embedding service to docker-compose.yml for easy local deployment, exposing port 8100 and using the new ARM64 image. (docker/docker-compose.yml)

Production QA Engine Integration

• Introduced a new QAEngine class that uses the Dockerized embedding service for inference and FAISS for vector search. It supports both single and batch query modes, handles remote embedding calls, and provides detailed logging and latency metrics. (src/embedding/qa_engine.py)