Amygda
Amygda, a Derby company, builds explainable AI for predictive maintenance, scoring the health of individual aircraft and airport equipment from maintenance and sensor data.
Founded in 2018 with up to ten staff, it combines fleet data from several manufacturers with historical fault patterns to isolate problems, and fuses sensors, logs and technician reports to monitor airport infrastructure. Its tools include a maintenance copilot and a time-series foundation model meant to replace one-off models for each asset type.
Its two flagship case studies, baggage systems at an unnamed major European airport and an airline fleet-health project, name no customer. The results it cites, such as 14 days' average warning and an 84 per cent detection rate, are the company's own.
No verified deployments yet
Amygda is profiled in the companies register, but no deployment has been verified against published evidence. Records are added as evidence is reviewed and graded.
Products & Solutions
5 productsRisk Score
AI system that processes existing operational data from CMMS systems, maintenance logs, fault codes, and work orders to assign risk scores to individual assets in a fleet.
Transport operators accumulate millions of maintenance records, fault codes, and CMMS logs that cannot be manually analysed in time to support operational decisions. The product addresses the inability to identify which specific assets in a fleet are at elevated risk of in-service failure before they depart on a service run, by automatically structuring messy operational data and surfacing risk patterns with sufficient advance warning for intervention.
- Ingests and auto-cleans raw CMMS exports, maintenance logs, fault codes, damage reports, and operational records without requiring new sensors
- NLP extraction reads both free-text human entries and machine-generated codes (e.g. ERR_4091_PWR, HVAC_FAULT:03)
- Deterministic processing with approximately 10% LLM usage, not a pure LLM system
- AI builds a weighted category hierarchy customised to the operator's asset type and industry, not a generic trained model
8 more features
- 0-to-1 risk score per asset displayed on a green/yellow/red colour scale for dispatcher use
- Every score is explainable: shows exactly which factors (recent maintenance events, historical patterns, operational context) contributed
- Real-time scoring as new CMMS entries are made
- Validated at 80% detection accuracy for at-risk assets with at least 60 minutes advance warning
- Scales from 1,600 to 18,000+ fault codes without proportional increase in processing time
- Supports longer-term forecasting 7 to 14 days ahead of developing risk
- Operator can edit, rename, split, or merge AI-generated categories with domain knowledge; AI re-maps immediately
- Works with CMMS platforms including Maximo and SAP
Technical specifications (7)
- Four-step processing pipeline: Ingest and Clean, Extract Meaning, Build Knowledge Map, Calculate Risk Scores
- Risk score output on 0-to-1 scale mapped from weighted multi-factor mathematical formula
- Weights learned from operator's own historical data, not pre-set
- NLP handles inconsistent formats, multiple languages, machine codes, and human-written descriptions
- Processes over 250 million operational records validated in production deployment
- Condensed 1,600+ fault codes into 57 actionable weighted categories in DB Regio validation
- Approximately 90% deterministic processing, approximately 10% LLM-based output generation
AI Engine
Self-service web-based data analysis platform for time-series equipment data. Designed for subject matter experts (SMEs) and engineers who do not have data science skills, enabling self-directed upload and analysis of transport equipment data without installation or implementation time.
Engineers and SMEs who need to investigate specific equipment problems or explore data opportunities lack access to expert-level data analysis tools without hiring data science specialists or waiting for IT implementation projects. The product provides a no-code, no-installation analysis capability accessible via free sign-up.
- Self-service data upload and analysis, no data science skills required
- Accepts time-series equipment data files up to 200 MB
- No credit card required, free sign-up available
- No installation or implementation time
4 more features
- Results downloadable in Excel format for internal use
- Multiple analytical models available (model list described as growing rapidly on the product page)
- Tested specifically for transport data
- Suitable for both individual engineers/SMEs and data science teams seeking rapid insight generation
Technical specifications (5)
- Web-based SaaS, no local installation required
- Maximum file size: 200 MB per upload
- Output format: Microsoft Excel (.xlsx)
- Designed for time-series equipment data
- Technical methodology details noted as coming soon on product page at time of research
AI Assist
AI-powered knowledge assistant engineered for engineering and maintenance documentation in transport operations. Uses large language models (LLMs) and document extraction algorithms to process technical documents containing text, tables, and images.
Operations and maintenance teams spend excessive time searching through large volumes of engineering documentation, manuals, maintenance records, procedures, to piece together troubleshooting guidance or retrieve relevant past resolutions. The product addresses this knowledge blindness by making unstructured engineering documentation searchable and queryable using AI, reducing dependency on the most experienced staff and accelerating troubleshooting resolution times.
- Multiple AI agents supporting different task types: maintenance manuals, fleet operations, troubleshooting, field services
- Best-in-class engineering document extraction for documents containing text, tables, and images
- Designed and trained specifically for transport operations (not a generic AI tool)
- Uses latest LLMs combined with proprietary extraction algorithms
4 more features
- Operates on self-hosted open source LLMs, no dependency on OpenAI or external training
- Analyses past resolutions from historical maintenance records to support troubleshooting
- Reduces skills gap by providing instant, accurate guidance to new staff
- Real-time anomaly detection and root cause analysis capability for airport baggage and security systems
Technical specifications (4)
- Utilises retrieval-augmented generation (RAG) optimised for engineering document processing in transport applications
- Supports extraction from documents combining text, tables, and images
- Self-hosted open source LLM deployment architecture
- Generates synthetic data from aerospace industry configurations for training where real data is limited
Foundation Model (Time Series)
Pre-trained, reusable AI model architecture built on time-series sensor data for transport asset health applications. Developed using unsupervised, semi-supervised, and self-supervised learning techniques to reduce dependency on large sets of well-annotated historical data.
Traditional AI solutions for asset maintenance require separate models to be built for each specific failure type or use case, taking 4 to 6 months per model and requiring large volumes of well-annotated historical data that operators frequently do not have. The foundation model architecture provides a flexible, reusable model that can be adapted to multiple use cases via transfer learning, removing the need to rebuild from scratch for each new application and enabling deployment even where annotated data is scarce.
- Single model architecture applicable to multiple use cases: smart maintenance, route optimisation, equipment life forecasting
- Transfer learning capability enables rapid adaptation to new use cases and asset types with minimal fine-tuning
- Unsupervised and semi-supervised training techniques reduce reliance on well-annotated historical datasets
- Self-supervised learning development ongoing at time of product documentation
6 more features
- Explainability algorithms layered on top of underlying model to provide interpretable output
- Combines human and AI input to create explainable and trustworthy models
- Supports synthetic data generation from aerospace industry configurations to supplement real data
- Network effect architecture: improves with each customer deployment, learning from broader fleet data
- Enables full fleet coverage for mixed-OEM fleets and novel, less-understood failure modes
- Reduces data engineering time from months to days for new use case deployment
Technical specifications (6)
- Time-series foundation model architecture trained on sensor data
- Transfer learning-based model using multiple deep learning architectures (architecture specifics not publicly disclosed)
- Unsupervised and semi-supervised training with self-supervised learning in development
- Synthetic data generation from aerospace industry configurations combined with real industrial data for training
- Explainability algorithms sitting alongside model for interpretability
- White paper available (download form required on product page)
Maintenance Copilot
AI-powered troubleshooting and root cause analysis (RCA) module integrated within the Amygda platform. Accepts plain-text input queries from maintenance and engineering teams, processes data from maintenance logs, fault codes, flight hours, pilot reports, and sensor data, and returns context-driven insights and answers.
Maintenance and engineering teams face time-intensive troubleshooting processes that require searching through extensive data, knowledge bases, and documentation to identify root causes of in-service incidents. The product accelerates this by enabling teams to query data in plain language and receive accurate, context-driven RCA recommendations, reducing reliance on the most experienced personnel.
- Plain-text natural language input for maintenance troubleshooting queries
- Context-driven root cause analysis (RCA) drawing from maintenance logs, flight hours, pilot reports, and sensor data
- Answers what-if scenarios based on operational data, not limited to historical query
- Remaining useful life (RUL) calculation for assets, stated as a unique feature beyond standard LLM capability
5 more features
- Fine-tuned specifically for transport and MRO operations (not a generic AI tool)
- Augments existing maintenance knowledge bases
- Faster troubleshooting with reduced response times
- Resolution derived from past maintenance actions
- Deployed across airlines, helicopters, and rail verticals per offering pages
Technical specifications (4)
- Generative AI (GenAI) architecture with transport-specific logic layer
- Accepts input from maintenance logs, fault codes, flight hours, pilot reports, and sensor data
- RUL computation logic built in as a distinguishing capability versus general LLMs
- Integrated within the broader Amygda platform, not sold as standalone product per available evidence
Profile data from the Aerostack companies register; deployment records from the verified register. Evidence grades: T1 first-party confirmation, T2 credible reporting. See Our approach for how records are graded.