Back to projects
Backend / Data RetrievalProduction buildMachine Learning Engineer

Knowledge Graph RAG (ToPWR)

FastAPI + Neo4j + LangGraph + MCP workflow that turns natural language into Cypher, deployed as a 4-service Docker stack.

A university data assistant built at Solvro to help students and staff navigate institutional information faster. The system combines a React chat UI, a FastAPI application layer, an MCP server, and a Neo4j knowledge graph so users can ask natural-language questions and retrieve structured university data.

Converted user questions into Cypher-backed graph retrieval through a FastAPI backend and an MCP server

Designed a LangGraph pipeline with LLM guardrails for query routing, Cypher generation, and response orchestration

Built a Prefect ETL pipeline ingesting PDF/text documents into the graph via OCR, extraction, and LLM-generated Cypher

Impact

Natural language mapped to graph-backed university data, in production at Solvro

Role

Machine Learning Engineer

Timeline

2025

Key tags

FastAPINeo4jRAG

Problem

University information is scattered across departments, pages, and systems, which makes discovery slow for students and staff.

Solution

Built a knowledge-graph retrieval workflow that checks query relevance, generates Cypher, executes graph retrieval, and returns answers through an application API.

Architecture

A React frontend, FastAPI backend, FastMCP server, and Neo4j run as a 4-service Docker stack. LangGraph orchestrates retrieval with guardrails, while a Prefect ETL flow processes documents and populates the knowledge graph.

Challenges

  • Modeling university entities and relationships cleanly in a graph schema
  • Handling ambiguous user intent without unsafe or misleading queries
  • Keeping retrieval tied to structured data while preserving conversational UX

Technology stack

PythonFastAPILangGraphLangChainFastMCPNeo4jPrefectDocker

Continue exploring

More case studies

Backend / AI Platform

Lucid - AI Study App

Helps students turn lecture screenshots and notes into searchable study material, flashcards, and quizzes.

Computer Vision / 3D

NeuroPark - 3D Vehicle Detection

Reconstructs 3D vehicle scenes from a single traffic-camera image, validated against LiDAR.

Backend / RAG Prototype

Medical Chatbot RAG

Provides educational health Q&A from retrieved source content instead of free-form LLM answers.