Car Diagnostics
AI diagnostics illustration

How AI Explains Trouble Codes: 8 Agents Under the Hood

How Car Diagnostics AI works: eight specialised agents analyse DTC codes, predict failures, generate reports, and answer questions — inside a microservice architecture.

AI-generated 4 min read

Car Diagnostics Platform uses 8 specialised AI agents for vehicle fault analysis. Each agent handles a narrow domain: one reads DTC codes from the ELM327 adapter, another analyses them in the context of the specific car make, another predicts future failures. This is not a single general-purpose assistant — it is a team of experts running on a microservice architecture, communicating through an event bus.

Why 8 agents instead of one general model?

Specialisation matters in automotive diagnostics. A general-purpose language model can explain that code P0420 means “catalytic converter fault”, but it cannot account for diagnostic protocol differences, UDS (ISO 14229) service procedures, ECU timing constraints, or adapter compatibility. Each detail requires domain-specific logic.

The 8-agent architecture delivers four concrete advantages:

  • Narrow expertise — each agent has focused knowledge of one domain (OBD-II protocols, SAE J2012 code definitions, brand-specific failure statistics)
  • Parallel execution — DTC scanning and failure prediction run concurrently without waiting for each other
  • Fault isolation — if one agent fails to respond, the remaining 7 continue working
  • Auditable output — you always know which agent produced which conclusion

The total platform DTC library contains 1947 generic SAE J2012 codes, and every analysis draws on this knowledge base.

Which agents work under the hood?

The 8 agents split into three functional groups:

Diagnostic core:

  1. Diagnostic Agent — the coordinator: receives DTC codes, distributes tasks to other agents, and assembles the final response
  2. OBD Reader Agent — handles all ELM327 adapter communication: protocol initialisation, PID polling, code reading through Mode 03 (stored), Mode 07 (pending), and Mode 0A (permanent)
  3. DTC Analyzer Agent — explains each code in the context of the specific make, model, year, and prior session history for that vehicle

Prediction and reporting:

  1. Predictor Agent — based on DTC history, identifies failure risk factors and estimates which systems may need attention over the near-term driving horizon
  2. Report Generator Agent — produces a structured diagnostic report in Markdown or PDF format
  3. Recommendation Agent — suggests a repair sequence: what needs immediate replacement versus what can wait

User interaction:

  1. Chat Agent — answers follow-up questions in the chat interface, retaining context from previous diagnostic sessions
  2. Admin Agent — internal service for system monitoring, logging, and caching

How do the agents communicate with each other?

The agent architecture uses an event-driven model over an event bus — each agent subscribes to specific message queues and publishes results asynchronously:

DTC scan → OBD Reader Agent → Diagnostic Agent
                                   ├── DTC Analyzer Agent (explanation)
                                   ├── Predictor Agent (forecast)
                                   ├── Recommendation Agent (repair plan)
                                   └── Report Generator Agent (document)
Chat Agent ←→ Diagnostic Agent (session context)
Admin Agent ← monitoring all queues

When you start a scan, the OBD Reader Agent polls the car’s ECUs through the ELM327 adapter. The Diagnostic Agent receives the codes and dispatches them in parallel to the DTC Analyzer, Predictor, and Recommendation agents. Results arrive asynchronously — you see output as each agent finishes. A full architecture overview is on the How It Works page.

A key feature: the system remembers previous sessions. If you scanned the same car a month ago, the DTC Analyzer compares the new codes against history and shows trends — new faults, resolved codes, or recurring patterns.

Is it safe to send car data to the AI?

Data safety is a platform priority. Here is how your information is protected:

  • Anonymisation — DTC codes and PID data are transmitted without the VIN (Vehicle Identification Number) or GPS location
  • Encryption — all connections between the app and servers use TLS encryption
  • Authentication — accounts are secured with time-limited access tokens
  • Logging — all agent requests are logged to a centralized logging system for audit purposes without retaining personal data
  • Guest mode — you can test diagnostics without creating an account

The platform does not sell or share vehicle fault data with third parties.

Agent deployment tiers. Not all 8 agents run at the same performance level. Agents that perform real-time OBD communication (OBD Reader, Diagnostic core) run on lower-latency infrastructure because they handle time-sensitive ELM327 polling. The Predictor and Report Generator agents, which perform heavier computation, can run on more powerful but slightly higher-latency nodes. The Chat Agent for user conversations runs on a mid-range tier that balances response speed and cost.

For more details, see the platform capabilities and the usage overview documentation.

The 8-agent system is designed to extract the maximum useful information from a single scan — from raw trouble codes to a mechanic-ready report — while keeping your data private and the process transparent.