TechCorp RAG Assistant

An end-to-end Retrieval-Augmented Generation (RAG) assistant built in Python that ingests corporate policy and product documentation to generate grounded answers with source citations.

Experience Gained

• Vector Retrieval Pipeline: Designed a vector retrieval pipeline using OpenRouter (text-embedding-3-small) for vector embeddings and Neon PostgreSQL (pgvector) for cosine similarity search.
• Low-Latency LLM Inference: Integrated Groq Inference API for low-latency completions, implementing automated prompt augmentation, confidence scoring, and fallback handling for failed API calls.
• Containerization & CI/CD: Containerized the application using Docker and GHCR CI/CD workflows, optimizing cloud deployment on Render with minimal memory footprint.
• Grounded Information Retrieval: Implemented strict document retrieval strategies and prompt constraints to eliminate hallucinations and deliver accurate, cited responses from corporate documents.

Technical Architecture

• Backend: Developed with Python and Flask to handle ingestion pipelines, querying APIs, and prompt orchestration.
• Vector Database: Neon PostgreSQL leveraged with the pgvector extension for high-performance vector similarity search.
• AI & Embeddings: Embeddings powered by OpenRouter (text-embedding-3-small) and LLM generation powered by Groq API.
• DevOps & Cloud: Packaged with Docker, versioned and published via GitHub Container Registry (GHCR) workflows, and hosted on Render.
• Frontend: Lightweight interface styled with HTML, CSS, and JavaScript for responsive query interaction.