Approved publication retrieval
Designed for maintenance questions that should be answered from current manuals, procedures, task cards, and local references.
Maintenance Control
Voice-first retrieval for aviation maintenance teams.
A private MVP build for approved-publication retrieval, voice query, and source-cited maintenance guidance on phone, web, and wearable surfaces.
Private buildWeb + iOS companionActive MVP
Operational shape
How the project supports decisions, context, and execution.
- Primary use
- Approved publication lookup
- AI role
- Grounded answers and voice query
- Best fit
- Frontline maintenance support
- Domain support
- Aviation SME review
What it is
Maintenance work is physical, procedural, and time-sensitive. Asking technicians to stop, search, type, and translate a manual into the next action creates friction at exactly the wrong moment.
Maintenance Control is built around voice-first access to approved publications. The current private build includes a Next.js web console, an Expo iOS companion, shared contracts, publication ingestion, parser review, revision-aware retrieval, persisted sessions, media upload plumbing, and benchmark tooling.
The product is not trying to authorize maintenance actions. It helps maintainers find the right source-backed answer faster, keeps citations visible, and leaves a record that supervisors and admins can review.
Implementation evidence
The project is positioned as maintainer workflow software, not a generic chatbot wrapped around technical manuals.
- Private monorepo with a Next.js web/admin app, Expo iOS companion, shared contracts, schema-backed API clients, and shared environment contracts.
- Controlled publication flow: draft upload, parser review, metadata confirmation, current revision activation, and maintenance-context binding.
- Grounded query path that scopes retrieval to the selected aircraft or component context and cites approved source chunks.
- Persisted query sessions, retrieved chunk hits, generated answers, media references, diagnostics, and audit events.
- Benchmark commands for retrieval, parser readiness, Super King Air demo readiness, and voice scenario lifecycle checks.
- Aviation SME review can be supported by Corrie Mays, a former Marine Corps aviator and Naval Flight Officer (NFO).
Current product architecture
The active build separates controlled content review, frontline query, and audit records so source grounding remains visible across web, phone, and wearable surfaces.
Rendering architecture diagram...
Maintenance Control keeps approved publications, scoped retrieval, voice input, answer composition, and audit records in separate parts of the system.
App example
Screenshots from the current mobile companion show source readiness, an answered maintenance query, and hands-free voice capture.
Source readiness
Shows the phone companion with a Super King Air B300/B300C context, approved sources ready, and glasses connection state visible.
Answer state
Shows the mobile answer surface after a maintenance query. Citation and source cards are part of the answer payload and companion flow.
Voice capture
Shows the ready state, voice command examples, and the hands-free voice entry point on the mobile companion.
Technical proof points
The useful evidence is in the implementation surfaces and the safety checks around retrieval, not in a polished audio demo.
Controlled sources
Publication upload and activation flows keep reviewed revisions separate from draft or superseded content.
Scoped retrieval
Queries filter to current approved revisions for the selected maintenance context before lexical and vector retrieval.
Citation behavior
Answer payloads carry citations, source excerpts, related sections, support level, and abstention state.
Voice lifecycle
Recorded voice, transcription, fallback behavior, and Realtime session seams are covered by scenario tests.
Audit records
Sessions preserve transcript, retrieved chunk hits, citation order, excerpt offsets, answer text, user, device, timestamp, and diagnostics.
Evaluation gates
Retrieval benchmarks track Top-1, Top-3, citation correctness, abstain behavior, and unsafe wrong-manual cases.
How it's used
These are the practical workflows and constraints the project is built around.
Voice-first companion
Uses phone and wearable surfaces for short questions, answer playback, media context, and source review when desktop software is the wrong surface.
Admin and review loop
Gives admins a place to manage publication intake, review parser output, activate revisions, and inspect query/session records.
Why this matters for operational AI
The hard part is not speaking an answer through a headset. The hard part is making sure the answer comes from the right approved source, for the right aircraft context, with enough citation and session history for a supervisor or admin to review later.
Status
Maintenance Control is an active private MVP build. The public page shows current implementation evidence; deeper repo access, transcript-based demos, or live walkthroughs can be shared with appropriate reviewers.
- The public view avoids exposing private repository details or operational data.
- Audio does not need to be the proof. The stronger proof is the state transition: spoken question, transcript, scoped retrieval, citations, answer, and audit trail.
More public projects
Different surfaces, same focus on context, coordination, review, and execution. This is a selected public view, not a complete project history.
AlignTrue
AlignTrue is an open-source reference architecture for supervised AI systems that need durable receipts, replayable state, governed actions, and fenced side effects.
GlassAlpha
GlassAlpha is an open-source toolkit for deterministic audit reports for tabular ML models, with fairness, explainability, calibration, robustness, and reproducibility metadata in one local workflow.
Ready to build something real?
Main Character helps turn complex product, workflow, and architecture problems into systems that work under real constraints.