HEYEJ

Aircraft maintenance teams spend a significant portion of their time on documentation — not wrenching, but searching. Troubleshooting a complex avionics fault or a recurring mechanical issue means manually combing through thousands of pages of manufacturer maintenance manuals, cross-referencing service bulletins, and leaning on the accumulated knowledge of senior engineers who may not always be available.

The consequences of this process are real: extended aircraft downtime, inconsistent diagnoses across shifts, and a fragile reliance on tribal knowledge that walks out the door when experienced technicians retire. The organization needed a better system — one that could surface the right answer from the right manual quickly, and that could learn from every case it processed.

The requirements set a high bar for AI trustworthiness:

• AI-powered diagnostics grounded in actual manufacturer manuals — not generalized knowledge that could hallucinate part numbers or procedures
• Mandatory citation of source manual sections for every recommendation, so technicians can verify against the original document
• Case-based troubleshooting history that builds an organizational knowledge base over time
• Multi-turn conversational interface that guides technicians through symptom isolation to diagnosis
• Multi-tenant architecture for different airlines and MRO organizations, each with isolated data and configuration
• Role-based access control distinguishing technicians, supervisors, and administrators

The citation requirement was non-negotiable. In a safety-critical industry, an AI that cannot show its work is an AI that cannot be trusted.

OneChair built the complete platform in approximately 5 hours. The architecture centers on a RAG (Retrieval-Augmented Generation) system that grounds every AI response in ingested maintenance manual content rather than general model knowledge.

Maintenance manuals are uploaded and processed asynchronously via BullMQ job queues. Each document is chunked, embedded using OpenAI's embedding models, and stored in PostgreSQL with the pgvector extension for semantic similarity search. When a technician describes a fault, the system retrieves the most relevant manual sections and passes them as grounded context to the language model — ensuring every response is anchored to verified source material.

Citation compliance is enforced at the architecture level, not as a post-processing step. The prompt structure requires the model to reference specific manual sections, and the response format includes traceable links back to the original document pages. Technicians can open the cited section directly from the diagnostic result.

The diagnostic workflow is structured as a guided conversation: manufacturer selection, aircraft type, affected system, problem description, and priority level. Each step narrows the retrieval scope, improving both accuracy and response relevance. Multi-turn conversation allows technicians to refine the diagnosis — asking follow-up questions, ruling out causes, and drilling deeper into specific subsystems.

As cases are resolved, their outcomes are logged and indexed. The system builds an organizational case history that supplements manual-based retrieval with real-world precedent from the same fleet and operating environment. Multi-tenant isolation ensures each organization's case data and manual library remain completely separate.

The platform was fully functional in approximately 5 hours — a complete enterprise diagnostic system delivered in a single focused session. Every capability in the original specification was implemented, including the citation requirement that most AI solutions struggle to satisfy cleanly.

• ~5 hour build time from specification to working, multi-tenant platform
• 11 screens delivered covering case management, the full diagnostic workflow, user management, aircraft configuration, and admin controls
• 100% citation compliance — every AI diagnostic response traceable to specific maintenance manual sections
• RAG architecture with pgvector semantic search across all ingested manual content
• Multi-tenant from day one: isolated data, users, and configuration per organization
• Asynchronous manual ingestion via BullMQ — large documents processed in the background without blocking the interface

• For aviation and industrial sectors: AI diagnostics are only trustworthy when grounded in verified source material. RAG with mandatory citations solves the hallucination problem that blocks AI adoption in safety-critical industries — not by filtering outputs after the fact, but by constraining the model's inputs from the start.
• For enterprise teams: Multi-tenant architecture does not have to be an afterthought bolted on after launch. Building tenant isolation in from the data layer means the platform is enterprise-ready on day one, not after a costly retrofit.
• For knowledge management: When experienced technicians retire, their diagnostic patterns no longer have to leave with them. A case history system turns years of troubleshooting experience into institutional knowledge that improves every future diagnosis.