A production-ready, containerized Multimodal Hybrid Retrieval-Augmented Generation (RAG) system built with an event-driven state machine architecture to query complex academic and research papers.
Multimodal Ingestion Pipeline: Ingests raw PDFs via Unstructured parsing, separating text, tables, and images. Generates dense vision-based captions for images before ingestion.
Hybrid Indexed Vector Store: Generates dual dense (HuggingFace sentence-transformers) + sparse (FastEmbed) context embeddings, indexing them into a high-performance Qdrant cluster.
LangGraph Orchestration: Replaces monolithic LLM flows with a stateful START → retrieve_node → generate_node → END micro-orchestration engine powered by Groq (Llama-3.3-70B).
End-to-End Evaluation: Integrated with the RAGAS framework to programmatically compute production telemetry across four metrics: Faithfulness, Answer Relevancy, Context Precision, and Context Recall.
FastAPI Core Service: Served via fully asynchronous routing layers optimized with automated /health liveness checks.
Enterprise CI/CD Workflow: Configured with GitHub Actions automation for inline Flake8 linting and multi-stage Docker builds triggered on every branch commit.
Pull and run this containerized microservice environment locally:
# Pull the production-ready image
docker pull amnakhalid123/arxivagent-rag:latest
# Run the API microservice container
docker run -p 8000:8000 --env-file .env amnakhalid123/arxivagent-rag:latest