AI · 2024 - 2025

DevGuardian AI - Multi-Agent Code Review Platform

A virtual senior engineering team of role-based AI agents, an Architect, a QA reviewer and a Security reviewer, that reads every code change, backed by a retrieval layer built for code: code-aware chunks, a dependency graph on Neo4j and hybrid search.

Start a similar project

DevGuardian AI - Multi-Agent Code Review Platform screenshot

Outcomes

Three role-based review agents, each with its own mandate, on Microsoft Agent Framework and Azure OpenAI
Retrieval built for code: class- and method-level chunks, GraphRAG on Neo4j, hybrid semantic, symbolic and graph search
Incremental re-indexing keyed to git commits, so a review never re-embeds the whole repository
Licensed to Valco AI and customised for its internal development team

Overview

DevGuardian AI is a code review platform Momentaj designed and built in 2024 and 2025 as a product of its own. A virtual senior engineering team of AI agents reads every change: an Architect, a QA reviewer and a Security reviewer, each with its own mandate. Underneath them sits a retrieval architecture designed for code rather than for documents.

The Challenge

Architecture governance doesn't scale the way engineering teams do. The senior engineers who can catch a bad architectural decision, a security gap or a quietly growing pile of technical debt are the same people who are too busy to review every pull request that needs it. Their judgment can't be written into a linter or a style guide. It depends on knowing how a change ripples through the codebase's dependency structure, and a diff alone doesn't show that.

What We Built

Three role-based agents review each change on ASP.NET Core and Blazor, with Microsoft Agent Framework and Azure OpenAI behind them. The retrieval layer is the core of the platform. Source is parsed with Roslyn into chunks at the class, method, interface and SQL-statement level, so each chunk is a unit of code with meaning, where generic RAG systems split by token count. The codebase's dependency graph is modelled in Neo4j, so an agent can reason about what else breaks when a piece of code changes, not only what looks similar. The index is keyed to git commit hashes through LibGit2Sharp and updates only what a commit changed. Semantic, symbolic and graph signals are combined for every query. Vectors live in SQLite, Hangfire runs the background jobs, and every repository builds up a Project Memory Bank: a persistent record of technical debt, architectural violations and review history, so each review starts from everything the system has already learned about that codebase.

Results

A working multi-agent review platform, licensed once, to Valco AI, with changes made for its internal development team. DevGuardian AI began as a startup idea and is not developed further or sold today, but its retrieval design, with code-aware chunking, GraphRAG and incremental refresh, is the pattern we now apply to every knowledge system we build on structured material.