Job
Description
At Momenta, we're committed to creating a safer digital world by protecting individuals and businesses from voice-based fraud and scams. Through innovative AI technology and community collaboration, we're building a future where communication is secure and trustworthy. The Role Key Responsibilities Design and Deliver Core Detection Pipeline Lead the development of a robust, modular deepfake detection pipeline capable of ingesting, processing, and classifying real-time audio streams with high accuracy and low latency. Architect the system to operate under telecom-grade conditions with configurable interfaces and scalable deployment strategies. Model Strategy, Development, and Optimization Own the experimentation and refinement of state-of-the-art deep learning models for voice fraud detection. Evaluate multiple model families, benchmark performance across datasets, and strategically select or ensemble models that balance precision, robustness, and compute efficiency for real-world deployment. Latency-Conscious Production Readiness Ensure the entire detection stack meets strict performance targets, including sub-20ms inference latency. Apply industry best practices in model compression, preprocessing optimization, and system-level integration to support high-throughput inference on both CPU and GPU environments. Evaluation Framework and Continuous Testing Design and implement a comprehensive evaluation suite to validate model accuracy, false positive rates, and environmental robustness. Conduct rigorous testing across domains, including cross-corpus validation, telephony channel effects, adversarial scenarios, and environmental noise conditions. Deployment Engineering and API Integration Deliver a fully containerized, production-ready inference service with REST/gRPC endpoints. Build CI/CD pipelines, integration tests, and monitoring hooks to ensure system integrity, traceability, and ease of deployment across environments. Ideal Profile Technical Skills ML Frameworks: PyTorch, TensorFlow, ONNX, OpenVINO, TorchScript Audio Libraries: Librosa, Torchaudio, FFmpeg Model Development: CNNs, Transformers, Wav2Vec/WavLM, AASIST, RawNet Signal Processing: VAD, noise reduction, band-pass filtering, codec simulation Optimization: Quantization, pruning, GPU acceleration DevOps: Git, Docker, CI/CD, FastAPI or Flask, REST/gRPC Preferred Experience Prior work on audio deepfake detection or telephony speech processing Experience with real-time ML model deployment Understanding of adversarial robustness and domain adaptation Familiarity with call center environments or telecom-grade constraints What's on Offer? Excellent career development opportunities Leadership Role Opportunity to make a positive impact Show more Show less
