𝗘𝗠𝗘𝗡𝗗𝗔𝗧𝗢𝗥

AI-Powered, Multi-Model Consensus Code Review Platform

Emendator sends your code to three independent LLMs in parallel, merges their findings into a single high-confidence review, and gives you back a scored, annotated, and improved version of your code — through a clean, editor-grade web interface.

---

📖 Overview

Emendator is a full-stack SaaS application that automates code review using a multi-model AI consensus pipeline. Instead of relying on a single LLM (which can hallucinate or miss issues), Emendator queries three different models simultaneously through OpenRouter, then runs their outputs through a custom consensus engine that deduplicates findings, resolves conflicting suggestions, and scores overall confidence — producing one trustworthy, well-reasoned review per submission.

Users can sign up (email/password, Google, or GitHub), paste or write code in a Monaco-powered editor, request a review, and receive:

• An overall quality score
• Categorized issues (bugs, security, performance, style)
• Concrete suggestions and line-level fixes
• An AI-improved version of the code with a side-by-side diff viewer
• A saved history of past reviews they can revisit or edit

---

⚙️ How It Works

┌─────────────┐      ┌──────────────┐       ┌───────────────────────────────────────┐
│   React     │ ──▶ │  Express      │ ──▶  │        OpenRouter Review Pipeline     │
│  Frontend   │ HTTP │   Backend     │      │                                       │
│(Vite/Tailwind)     │  (REST API)   │      │  ┌────────┐ ┌───────────┐ ┌──────────┐│
└─────────────┘      └──────┬───────┘       │  │ Gemma  │ │ Nemotron  │ │ GPT-OSS  ││
                             │              │  └───┬────┘ └─────┬─────┘ └────┬─────┘│
                     ┌───────▼────────┐     │      └─────────────┼────────────┘     │
                     │  PostgreSQL    │     │           (run in parallel)           │
                     │  via Prisma ORM│     │                    ▼                  │
                     └────────────────┘     │        Consensus Engine (dedupe,      │
                                            │     conflict resolution, scoring)     │
                                            └───────────────────────────────────────┘

Step by step:

1. Authentication – Users register/login with email + password (bcrypt-hashed, JWT access + refresh tokens in HTTP-only cookies) or via Google/GitHub OAuth.
2. Submitting code – From the Review page, the user pastes code into a Monaco Editor and selects a language (18+ supported languages).
3. Parallel AI dispatch – The backend fans the code out to three independent models at once via Promise.allSettled, so total latency is bound by the slowest model, not the sum of all three.
4. Consensus building – The consensus.engine.js module:
   • Deduplicates overlapping findings using Levenshtein similarity (short strings) and trigram Jaccard similarity (longer strings)
   • Detects and resolves conflicting recommendations, keeping the majority view
   • Aggregates individual model scores into one confidence-weighted score
   • Selects the strongest candidate for the "improved code" output
5. Persistence – The final review (score, issues, suggestions, line fixes, improved code, metadata) is saved to PostgreSQL via Prisma, scoped to the authenticated user.
6. Review & History UI – Results are rendered with a score ring, categorized issue cards, and a diff editor comparing original vs. AI-improved code. Past reviews are listed on the History page and can be reopened or edited.
7. Resilience – If a model has no API key configured or fails, it's automatically excluded from the run (via Promise.allSettled) rather than failing the whole request; rate limiting and caching protect the AI layer from abuse.

---

🧰 Tech Stack

Frontend

Technology	Purpose
React 18	UI library
Vite	Build tool / dev server
Tailwind CSS	Utility-first styling
Zustand	Lightweight state management (auth & review stores)
React Router DOM	Client-side routing & protected routes
Monaco Editor (@monaco-editor/react)	In-browser code editor + diff viewer
Axios	HTTP client
React Hot Toast	Toast notifications
Lucide React	Icon set

Backend

Technology	Purpose
Node.js + Express	REST API server
Prisma ORM	Type-safe database access & migrations
PostgreSQL	Primary data store (Users, Reviews)
JWT (jsonwebtoken)	Access + refresh token authentication
bcryptjs	Password hashing
Passport-style OAuth (Google & GitHub)	Social login
express-validator	Request payload validation
express-rate-limit	Global & AI-specific rate limiting
Helmet, CORS, cookie-parser, morgan	Security headers, cross-origin handling, request logging
Winston + winston-daily-rotate-file	Structured, rotating application logs

AI / Review Pipeline

Technology	Purpose
OpenRouter API	Unified gateway to multiple LLMs
Gemma, Nemotron, GPT-OSS (via OpenRouter)	The three models queried in parallel for each review
Custom Consensus Engine	Deduplicates & merges multi-model output into one verdict
Confidence Scorer	Produces an aggregated confidence/quality score
In-memory Review Cache	Avoids re-querying models for identical code submissions

Infrastructure

Technology	Purpose
Docker & Docker Compose	Containerized Postgres, backend, and frontend services
Nginx (frontend container)	Serves the built React app

---

📁 Project Structure

emendator/
├── backend/
│   ├── src/
│   │   ├── config/            # DB, logger, OAuth, OpenRouter/OpenAI config
│   │   ├── controllers/       # auth, review, history controllers
│   │   ├── middleware/        # auth, rate limiting, validation, error handling
│   │   ├── routes/            # /api/auth, /api/review, /api/history
│   │   ├── services/
│   │   │   ├── openrouter/
│   │   │   │   ├── models/        # gemma, nemotron, gptoss clients
│   │   │   │   ├── consensus/     # multi-model consensus engine
│   │   │   │   ├── scoring/       # confidence scorer
│   │   │   │   ├── prompts/       # system/user prompt builders
│   │   │   │   └── cache/         # review response cache
│   │   │   ├── ai.service.js
│   │   │   ├── auth.service.js
│   │   │   ├── review.service.js
│   │   │   └── history.service.js
│   │   └── utils/              # JWT helpers, response formatting
│   └── prisma/schema.prisma    # User & Review data models
│
├── frontend/
│   ├── src/
│   │   ├── pages/               # Login, Signup, Dashboard, Review, History, Edit, OAuth callback
│   │   ├── components/          # CodeEditor, DiffEditor, ReviewResult, ScoreRing, Navigation, etc.
│   │   ├── store/                # Zustand auth & review stores
│   │   └── api/                  # Axios API clients
│
└── docker-compose.yml           # Postgres + backend + frontend orchestration

---

🚀 Getting Started

Prerequisites

• Node.js 18+
• PostgreSQL 15 (or use the provided Docker service)
• OpenRouter API keys (free tier available at openrouter.ai)
• (Optional) Google & GitHub OAuth app credentials

1. Clone & configure environment

git clone <your-repo-url>
cd emendator
cp .env.example .env   # then fill in your own secrets — see note below

⚠️ Security note: Never commit real API keys, database passwords, JWT secrets, or OAuth client secrets to version control. Generate fresh, unique values for JWT_SECRET, JWT_REFRESH_SECRET, POSTGRES_PASSWORD, and your OpenRouter/Google/GitHub credentials, and keep your .env file out of git (it's already listed in .gitignore).

2. Run with Docker (recommended)

docker compose up --build

This starts:

• postgres on :5432
• backend (Express API) on :5000
• frontend (Vite build served via Nginx) on :5173

3. Run manually (without Docker)

# Backend
cd backend
npm install
npx prisma migrate dev
npm run dev            # http://localhost:5000

# Frontend
cd frontend
npm install
npm run dev             # http://localhost:5173

---

🔑 Key API Endpoints

Method	Endpoint	Description
POST	/api/auth/register	Register with email/password
POST	/api/auth/login	Login and receive JWT tokens
GET	/api/auth/google, /api/auth/github	Start OAuth flow
GET	/api/auth/me	Get current authenticated user
POST	/api/review	Submit code for multi-model AI review
GET	/api/review/:id	Get a single review
GET	/api/history	List a user's review history
DELETE	/api/history/:id	Delete a review

---

🗄️ Data Model (Prisma)

• User — id, email, name, password (nullable for OAuth users), avatar, OAuth provider/providerId, related Review[]
• Review — id, userId, code, language, score, issues, suggestions, lineFixes, improvedCode, metadata (JSON), timestamps

---

🛡️ Security Features

• Passwords hashed with bcrypt
• JWT access + refresh token rotation via HTTP-only cookies
• Helmet security headers & strict CORS policy
• Rate limiting on global traffic and separately on AI-heavy endpoints
• Input validation on every mutating route via express-validator
• Centralized error handling middleware

---

📌 Notes

• The AI pipeline gracefully degrades: if one or two model API keys are missing or a model call fails, the review still completes using whichever models succeeded.
• Review results are cached in-memory to avoid redundant model calls for identical submissions.
• cre.py in the repo root is a developer utility script used to flatten the project into a single Markdown file (like the one this README was generated from) — it is not part of the running application.

---

Built with a Node.js/Express + React stack, PostgreSQL, and a multi-model AI consensus pipeline powered by OpenRouter.