Owned Retrieval Pipeline Architecture
Deep Learning RAG Interview Prep Agent
Pipeline Engineer | Retrieval Systems & LLM Workflow Orchestration
Project Summary
This project is a decision-ready retrieval-augmented study assistant for deep learning interview preparation. The system transforms source material into searchable knowledge, then routes user questions through an LLM workflow that retrieves relevant context and returns grounded, citation-aware responses.
My Role
I owned the pipeline engineering layer of the project. My responsibility was to turn raw study documents into a dependable retrieval system by connecting ingestion, vector storage, retrieval logic, and agent orchestration into one backend workflow the rest of the team could build on.
Key Contributions
1. Retrieval Pipeline Assembly
Engineered the core RAG backend across LangChain, LangGraph, and ChromaDB, creating the system path from uploaded document to retrieved context to final agent response.
2. Vector Store and Ingestion Flow
Implemented the ingestion and vector store workflow needed to initialize collections, process source material into searchable chunks, and make the corpus queryable for downstream LLM tasks.
3. Duplicate Detection and Query Reliability
Built duplicate-check and retrieval safeguards that reduced noisy re-ingestion and improved the consistency of query results, helping the agent stay grounded in the most relevant deep learning content.
4. Team Integration Support
Served as the backend integration point across prompt, UX, and QA workstreams by providing stable interfaces the team could use during testing, integration, and final demo preparation.
Technical Stack
PythonLangChainLangGraphChromaDBStreamlit- Retrieval-Augmented Generation (RAG)
Outcome
The finished application allowed users to upload deep learning study material, query a vectorized knowledge base, and receive grounded interview-style responses supported by retrieved source context. My contribution centered on building the retrieval backbone that made the end-to-end agent usable, testable, and demo-ready for a team setting.