jeff joji

work

On-device inference, the pipelines around it, and a couple of things shipped end to end.

01

experience

AptivSoftware Engineer2025/01 – Present

Engineered a conversational edge AI agent, showcased at CES 2026: low-latency ASR/TTS and quantized LLMs wired to AWS and Workday REST APIs for real-time, natural in-vehicle voice queries.

Accelerated on-device inference on Jetson Orin Nano and Qualcomm 8295p with TensorRT and C++ graph optimizations, built the MLOps fine-tuning and benchmarking platform that earned the Best Rookie Award, and orchestrated the Kubernetes and Terraform pipelines that ingest multi-terabyte sensor logs for scene-prediction training.