Loading Experience
Back to Experience

DRDO — Research Centre Imarat (RCI)Jul 2025 – Sep 2025Hyderabad, India

Responsibilities

  • Real-time Sign Board Classification and Translation (Python, OpenCV, TensorFlow, YOLO Nano): Built a real-time roadside sign board classification and translation system in Python, using lightweight quantized vision models on edge hardware to cut inference latency to under 15 ms per frame.
  • OCR & Multi-Class Script Recognition: Engineered OCR and script-recognition pipelines that detect multi-class local scripts with over 98% precision and recall under motion blur, low light, and physical distortion.
  • Sensor Processing & Trajectory Prediction Simulation (MATLAB, Simulink): Built a multi-sensor signal processing simulation in MATLAB and Simulink that ingests telemetry, radar tracking, and altimeter data across aerospace protocols, reaching sub-meter spatial RMSE in real-time 3D tracking.
  • Dynamic Target Tracking & Latency Reduction: Implemented machine-learning trajectory prediction for dynamic target tracking, cutting state-estimation latency by 45% while keeping detection probability above 99% under heavy noise.
  • Hardware Protocol Software (C, RS-422, MIL-STD-1553): Wrote C software that talks to hardware over RS-422 and MIL-STD-1553. Systematic testing and tight hardware-software integration cut communication errors by 35% and improved protocol compliance.
  • WinAPI Real-Time Telemetry Dashboard: Built a WinAPI dashboard that graphs incoming data in real time, eliminating manual log inspection, and overhauled data handling, synchronization, and error handling in the existing codebase.

Key Achievements

  • Cut edge vision inference latency to under 15 ms per frame using lightweight quantized YOLO Nano models on constrained hardware
  • Attained >98% precision and recall across multi-class local script recognition under motion blur, low light, and physical distortion
  • Reached sub-meter spatial RMSE in real-time 3D tracking by fusing telemetry, radar tracking, and altimeter data in MATLAB & Simulink
  • Reduced state-estimation latency by 45% with ML trajectory prediction while sustaining >99% detection probability in high-noise environments
  • Decreased communication errors by 35% and enhanced protocol compliance across RS-422 and MIL-STD-1553 buses
  • Eliminated manual log inspection for engineering teams by building a responsive WinAPI real-time data graphing interface

Technologies Used

PythonOpenCVTensorFlowYOLO NanoMATLABSimulinkCWinAPIRS-422MIL-STD-1553Sensor FusionEdge MLReal-Time Systems

Other Roles