AI Engineering (GPU-Accelerated Code Integration For Multimodal Systems) - Internship

Selected intern's day-to-day responsibilities include:

1\. Implement GPU-accelerated inference pipelines for person detection,

obstacle detection, and free-space estimation using optimized DL models.

2\. Convert and integrate pre-trained models (YOLOv8, DeepLab, etc.) for

facial emotion recognition and object tracking using TensorRT or ONNX.

3\. Build robust pipelines for real-time text detection (OCR) and audio

transcription (speech-to-text) using webcam and mic inputs on a laptop.

4\. Optimize latency and throughput of feature modules using micro-batching,

model quantization, and memory-efficient loading techniques.

5\. Develop asynchronous multi-threaded pipelines for concurrent audio and

vision tasks, enabling real-time response and resource sharing.

6\. Integrate lane detection and emergency condition detection into GPU-

executable modules, with fallback logic to CPU where needed.

7\. Create a unified deployment-ready build for multimodal inference on Linux-

based GPU systems (including simulated Jetson Nano setup on desktop GPU).

8\. Implement stream-based execution pipelines for live camera and audio feed

processing with minimal delay and max efficiency.

9\. Benchmark and compare performance between different model architectures

(e.g., MobileNet vs. ResNet, YOLOv5 vs. YOLOv8) across features.

10\. Integrate wake-word and voice command modules using lightweight ASR

models optimized for GPU or CPU deployment.

11\. Implement visual SLAM and mapping modules, simulate motion planning with

pre-recorded video or webcam input, and test with known datasets.

12\. Implement GPU-supported activity recognition and scene classification

using both audio and video cues (multi-modal input fusion).

13\. Collaborate with feature development interns to ensure seamless

integration and performance handoff of each module in the unified pipeline.

14\. Maintain documentation and profiling logs of inference time, GPU/CPU

usage, and module interdependencies.

15\. Assist in building testable demo apps for feature-level showcase and full

pipeline simulation on GPU-backed systems (desktop or cloud-based).