Telemetry — Deep Space Code Intelligence

A Mars-mission-themed AI code review, refactoring & architecture assistant.

Multi-provider LLM analysis (Gemini, Groq, local Ollama), a gamified review console, prompt-injection defenses, real-time telemetry, and a JWT-secured tunnel that injects AI-generated code straight into VS Code.

Project status — This is a personal / educational work-in-progress ("Mission Control"), not a production service. It is shared publicly as a portfolio and reference implementation. See Security & configuration before running it yourself.

---

Overview

Telemetry (internal codename CoderAI) reimagines the code-review experience as a deep-space mission console. You paste a snippet or point it at a local workspace, pick an AI "instrument", and the platform runs a structured review — an executive summary, actionable improvements, and a numeric quality score — while streaming everything back through a live, NASA-flight-deck-inspired UI.

It is a full monorepo spanning four surfaces:

Surface	Stack	Role
Backend API	.NET 8, Clean Architecture, EF Core, SignalR	Auth, agent orchestration, telemetry, real-time hub
Web console	Next.js 16, React 19, Tailwind 4	The Mission Control UI you see below
VS Code extension	TypeScript, SignalR client	"Ground Station" — receives AI code injections in-editor
CLI	Node.js	Connects a local workspace to the console for agent-mode analysis

---

Screenshots

Mission Control Dashboard

The command deck: pick an instrument, toggle Agent vs Assistant mode, and transmit code for analysis. XP / level progression lives in the top bar.

Assistant Mode — Code Review

Paste a snippet, choose a provider (Groq / Gemini / local), and run a full structured review.

Telemetry Metrics Console

A live vector HUD: session counts, agent-activity allocations, model-latency maps, security-shield diagnostics, and gateway-traffic ingestion — polled every 10s.

---

Features

• Multi-provider AI analysis — routes reviews across Google Gemini 2.5 Flash, Groq, and a local Ollama model (qwen2.5-coder) so you can trade cost, speed, and privacy per request.
• Agent & Assistant modes — Assistant reviews a pasted snippet; Agent connects to a local workspace (via the CLI) and reasons over real files with tools like explain, refactor, unit-test, security-audit, analyze, and rag-query.
• Structured, hallucination-resistant reviews — strict system prompts force an Executive Summary → Actionable Improvements → Score shape instead of free-form rambling.
• RAG engine — global and user-specific retrieval-augmented context built on vector embeddings for grounded answers.
• Gamification with anti-cheat — parses AI scores into an XP / Level system, backed by a SHA-256 code-hashing guard in the database to prevent XP farming from resubmitted code.
• Security shield — prompt-injection and homoglyph-obfuscation detection, JWT auth with silent refresh rotation, per-endpoint rate limiting, and API-key hashing for external clients.
• Real-time VS Code injection — a JWT-secured SignalR tunnel (CodeHub) pushes AI-generated code from the web straight into the editor.
• Live telemetry — a mission-control metrics console with latency maps, gateway traffic, and security diagnostics.
• Isolated code sandbox — a locked-down, network-less container for running untrusted code.

---

Architecture

CoderAI/
├── CoderAI.Domain/          # Entities & core domain (no dependencies)
├── CoderAI.Application/     # Interfaces, DTOs, use-case contracts
├── CoderAI.Infrastructure/  # EF Core, external integrations
├── CoderAI.Api/             # ASP.NET Core Web API + SignalR CodeHub
│   ├── Controllers/         #   Auth, CodeReview, Metrics, Gateway, Rewards, ...
│   ├── Hubs/                #   CodeHub — real-time editor tunnel
│   ├── Services/            #   Agent orchestration, AI providers, security
│   └── Middlewares/         #   API-key auth, request pipeline
├── CoderAI.Tests/           # xUnit test suite
├── frontend/                # Next.js 16 Mission Control web console
├── telemetry-extension/     # VS Code "Ground Station" extension
├── telemetry-cli/           # Local workspace bridge CLI
└── docker-compose.yml       # postgres, ollama, api, ui, sandbox

The backend follows Clean Architecture — dependencies point inward (Api → Application → Domain), with infrastructure and AI providers plugged in at the edges.

Tech stack: .NET 8, Entity Framework Core, ASP.NET Identity, JWT Bearer, SignalR, FluentValidation, Serilog · Next.js 16, React 19, TailwindCSS 4, Axios, @microsoft/signalr, react-markdown · PostgreSQL (prod) / SQLite (dev) · Docker Compose.

---

Quick start

Prerequisites

• .NET 8 SDK
• Node.js 20+
• (Optional) Docker — for the one-command stack
• (Optional) Ollama with qwen2.5-coder:7b — for local, offline AI

1. Backend (development uses SQLite — zero external services)

cd CoderAI.Api

# Provide a JWT signing key and an AI key via user-secrets (never commit these)
dotnet user-secrets set "JwtSettings:Secret" "a-random-string-at-least-32-characters-long"
dotnet user-secrets set "GeminiSettings:ApiKey" "YOUR_GEMINI_API_KEY"   # optional

dotnet run   # -> http://localhost:5109  (Swagger at /swagger)

2. Web console

cd frontend
cp .env.example .env.local     # points the UI at http://localhost:5109
npm install
npm run dev                    # -> http://localhost:3000

Open http://localhost:3000, request access, and log in.

3. Everything at once with Docker

cp .env.example .env           # fill in POSTGRES_PASSWORD, JWT_SECRET, GEMINI_API_KEY
docker compose up --build

This brings up PostgreSQL, Ollama, the API, the web UI, and the isolated sandbox on one network.

---

Security & configuration

No secrets are committed to this repository. All keys and credentials are supplied at runtime through environment variables, .NET user-secrets, or git-ignored files:

What	Where it goes	Template
JWT signing secret	JwtSettings__Secret env / user-secrets	—
Database connection	ConnectionStrings__DefaultConnection env	appsettings.Production.example.json
Gemini / Groq keys	GeminiSettings__ApiKey env / user-secrets	—
SMTP credentials	EmailSettings__* env / user-secrets	—
Docker stack secrets	root .env (git-ignored)	.env.example
Frontend API URLs	frontend/.env.local (git-ignored)	frontend/.env.example

appsettings.json ships with empty secret fields on purpose; in Development the app falls back to a clearly-labelled throwaway key, and in Production it refuses to start if a real JwtSettings__Secret is not provided.

Rotate any key that ever touched a real deployment before making a fork public.

---

Tests

dotnet test                          # backend (xUnit)
cd frontend && npx playwright test   # end-to-end UI (Playwright)

---

License

Released under the MIT License — see LICENSE.

_{Built with a love of clean architecture and a good Mars-landing metaphor.}