IDARA SAMUEL OSU — Machine Learning Engineer
Speech AI • Low-Latency Inference • Agentic Systems • Production MLOps
CORE STACK & TOOLING
Inference Engineering:
TensorRT-LLM vLLM SGLang CTranslate2 KV Caching Quantization In-Flight Batching
ML, NLP & Speech:
PyTorch Hugging Face LLMs Instruction Tuning NMT (Meta NLLB) TTS / STT Whisper Pipecat WebRTC / VAD
Agentic Frameworks:
LangGraph LangChain Multi-Agent Orchestration Tool / Function Calling
Cloud & MLOps Infrastructure:
AWS SageMaker Model Registry Multi-Model Endpoints Model Monitor Modal Docker CI/CD Supabase Python
TECHNICAL FOCUS & CAPABILITIES
|
Low-Latency Serving Stack Compiling open-weight LLMs, Whisper, NMT, and TTS/STT models via TensorRT-LLM and CTranslate2; implementing KV caching, quantization, and in-flight batching for sub-second execution. |
|
Full-Duplex Interactive Voice Architecting real-time speech-to-speech pipelines using Pipecat with WebRTC, VAD-driven turn-taking, and barge-in support for conversational voice agents. |
|
Agentic Orchestration Designing stateful multi-tool routing environments and multi-agent coordination architectures using LangChain and LangGraph with function calling. |
|
Cloud-Native MLOps & Production Pipelines Building automated SageMaker Pipelines (preprocess, train, evaluate, register), multi-model endpoints, serverless inference layers, and automated CI/CD deployments with Docker. |
|
Data Curation & Multilingual AI Architecting raw dataset collection pipelines across multi-thousand-hour speech and text corpora for low-resource and zero-digital-footprint languages. |
📬 Contact: idaraosusamuel@gmail.com | LinkedIn