Computer Scientist 2

Computer Scientist (preferably Level
2) – Need as soon as possible. Preferred Experience

• End-to-end neural speech-recognition systems using wav2vec 2.0, Whisper, and Deep Speech 2-style acoustic models.
• Sequence-to-sequence and speech models for multilingual and domain-specific applications.
• Experience with curating real and synthetic datasets for training, validation, and model comparison.
• Building datasets for repeatable benchmarking and regression testing.
• Designing benchmark suites measuring model accuracy, inference latency, package size, and release readiness.
• Designing hybrid recognition architectures that combine local processing with remote API calls.
• Applying language models and NLP techniques to correct likely recognition errors and improve contextual interpretation.
• Building scalable ETL and data-enrichment pipelines using e.g. Python, PySpark, Rust, Go, and FastAPI.
• Developing production ML systems using PyTorch, Tensor Flow, Keras, MLflow, ZenML, and related tools.
• Implementing artifact validation, deployment controls, and AI governance for production model releases.
• Applying anomaly detection, statistical modeling, and error analysis to identify operationally significant model failures.
• Rapid AI prototyping and translating research models into reliable, production-grade software pipelines.

• Design, implement, and maintain a centralized repository for CASPER training data, validation data, transcriptions, command labels, model outputs, and evaluation results.
• Provide full-stack, Dev Ops-oriented development across the training and validation environment, including front-end data-management interfaces, back-end services and APIs, databases, processing infrastructure, automated builds, deployment, monitoring, and operational support.
  4
• Develop aviation-specific model harness for LLM API integration
• Develop and train custom language models for simulated pilot speech and behaviors
• Build scalable ETL, labeling, and data-enrichment pipelines that map recorded and synthetic speech to CASPER commands, callsigns, fixes, digits, and other recognition targets.
• Maintain dataset, label, model, and command-schema versioning so training and evaluation data remain aligned as CASPER capabilities and phraseology mappings change.
• Curate representative training and held-out validation datasets covering different speakers, phraseology, audio conditions, callsign types, commands, and operational contexts.
• Automate end-to-end model training, replay, benchmarking, validation, and nightly or per-build regression testing of the CASPER recognizer and parser.
• Measure and report word error rate, callsign accuracy, digit accuracy, command and intent accuracy, precision and recall, latency, regressions, and release readiness.
• Use exploratory, hypothesis-driven, and experiment-based methods to identify, prototype, and evaluate alternative approaches for improving voice-recognition performance.
• Design controlled experiments and ablation studies comparing acoustic models, language models, contextual sources, preprocessing methods, confidence thresholds, and recognition-pipeline configurations.
• Develop, fine-tune, and evaluate ATC-specific speech-recognition models, including wav2vec 2.0, Whisper, Deep Speech 2-style models, sequence-to-sequence models, and other promising architectures.

Remote work may be permitted when consistent with project needs. On-site presence will be required as dictated by FAA project needs.

Masters or Phd in Computer Science or related field preferred