Portfolio

Detailed project portfolio and contributions @ Partridge Systems @ EASY AI

Realtime Driving Image/Video Processing Pipeline Integration

2025.09 - 2025.10

Data Engineer @ Partridge Systems

Built a real-time metadata extraction pipeline for ADAS/AD-related objects detected in driving images(batch) and videos. Leveraged NVIDIA DeepStream for GPU-accelerated inference and delivered extracted metadata to downstream services via FastAPI.

Tech Stack

PythonNVIDIA DeepStreamGStreamerYOLOFastAPIDocker

Key Highlights

  • •Designed end-to-end pipeline from video ingestion to structured metadata output using NVIDIA DeepStream
  • •Achieved real-time performance for ADAS/AD object detection and metadata extraction
  • •Exposed pipeline results through a FastAPI service and containerized the entire stack with Docker

De-Identification Implementation in ADAS/AD Validation Platform, Databahn

2025.10 - current

Computer Vision Engineer @ Partridge Systems

Implemented an automated de-identification pipeline for driving videos captured by vehicle front cameras. Detects and anonymizes pedestrian faces and license plates to comply with privacy regulations, enabling safe use of real-world driving data in the ADAS/AD validation platform.

Tech Stack

PythonNVIDIA DeepStreamGStreamerYOLOFastAPIDockerCUDA/C++

Key Highlights

  • •Automated detection and anonymization of pedestrian faces and license plates in front-camera driving footage
  • •Integrated de-identification component(a.k.a. Job Component) into the existing Databahn validation pipeline
  • •Enabled privacy-compliant use of real-world driving data for ADAS/AD model validation

Databahn Job Component Migration (C++ → Python)

2026.01 - current

Data Engineer @ Partridge Systems

Migrated a legacy C++-based job responsible for parsing CAN signals and GPS data to extract ADAS/AD-related metadata into a maintainable Python implementation, improving development velocity and cross-team accessibility.

Tech Stack

C++PythonCANDBCGPSMDF4 (mf4)

Key Highlights

  • •Re-implemented CAN signal and GPS parsing logic from C++ to Python with equivalent output fidelity
  • •Parsed industry-standard DBC and MDF4 (mf4) formats for ADAS/AD metadata extraction
  • •Improved code maintainability and reduced onboarding time for new team members