Hi everyone,
I wanted to share an update on a project I’ve been building called TunersX, along with its experimental AI layer COBALTGPT.
The goal is to build a high-integrity vehicle telemetry and diagnostics platform that captures and analyzes vehicle network data (CAN / OBD / J2534) while maintaining strong safety and compliance boundaries.
This project started as a system for my 2010 Chevrolet Cobalt SS Turbo (LNF / E69 ECM) but is evolving into a broader vehicle diagnostics and telemetry framework.
⸻
Project Concept
TunersX is designed around a trace-bundle workflow:
Capture → Decode → Verify → Report
The idea is to produce engineering-grade evidence bundles from vehicle data that can be inspected, replayed, and verified.
The platform captures CAN / OBD telemetry, decodes signals using DBC / PID maps, and generates structured reports like:
• Health overview
• Anomaly detection
• Timeline visualization
• Evidence bundles for review
The overall goal is reliable diagnostics and telemetry, not reverse-engineering or bypass tooling.
This architecture is described in the project charter as a vehicle telemetry toolkit with evidence-grade capture and reporting.
⸻
Core Architecture
Current architecture components include:
1. Capture Layer
Interfaces with vehicle networks using:
• CAN interfaces (CANable, SavvyCAN compatible)
• J2534 devices (initial focus: **RLink X3**)
• OBD-II transport
Data is timestamped and stored in trace bundles for analysis.
⸻
2. Decode Layer
Signals are decoded using:
• DBC definitions
• PID registries
• derived metrics
Decoded signals feed into:
• health metrics
• anomaly detection
• timeline analysis
⸻
3. Policy & Safety Layer
TunersX intentionally runs in PASSIVE mode by default.
Active operations require:
• explicit runtime authorization
• precondition checks (power, connection stability)
• policy allowlist
• full audit logging
This ensures the platform cannot perform sensitive operations without clear user approval and logging.
Important boundaries:
• no SecurityAccess implementation
• no immobilizer or theft bypass
• no OEM programming procedure replication
⸻
4. Evidence Bundle System
Every session generates a trace bundle containing:
• captured CAN frames
• decoded signals
• integrity manifests
• audit logs
• metadata
These bundles can be exported as a Reviewer Pack including reports and artifacts.
⸻
5. Dashboard + Analysis
The planned dashboard provides:
• bundle integrity status
• vehicle health overview (boost, IAT, knock retard, trims)
• event timeline
• anomaly queue
• export tools for reports
The design emphasizes data provenance, version tracking, and evidence pointers for each chart.
⸻
AI Layer — COBALTGPT
COBALTGPT is an experimental AI layer on top of TunersX.
It functions as a:
• diagnostic reasoning engine
• telemetry analysis assistant
• documentation system
Rather than controlling vehicle systems, it analyzes telemetry and produces engineering-style diagnostics reports.
⸻
Current Status
Working pieces:
• CAN capture prototypes
• DBC parser and decoding pipeline
• Python analysis tooling
• early dashboard concepts
• documentation and architecture charter
In progress:
• J2534 readiness validation layer
• evidence bundle packaging
• anomaly detection engine
• reviewer pack exports
• dashboard MVP
⸻
Research Goals
The broader research questions I’m exploring include:
• How to build **tamper-evident telemetry logs** for vehicle networks
• How to structure **trace bundles for reproducible diagnostics**
• Whether AI can assist with **vehicle fault analysis**
• How to safely orchestrate diagnostics workflows without bypassing OEM protections
⸻
Looking for Feedback On
I’d love feedback from anyone working in:
• vehicle telemetry
• CAN tooling
• automotive diagnostics
• AI-assisted engineering analysis
• open vehicle data standards
Specifically:
• best practices for **trace bundle formats**
• CAN capture integrity techniques
• anomaly detection strategies for vehicle telemetry
• dashboard UX for engineering diagnostics
⸻
Final Note
TunersX is being built with strong compliance boundaries and focuses on diagnostics, telemetry, and analysis, not bypass tooling.
The goal is to explore how modern software tooling and AI could improve vehicle diagnostics and engineering workflows.
If anyone is interested in discussing architecture ideas or telemetry tooling approaches, I’d love to connect.