Natesh Tyagi

Windows-on-Arm Camera

Camera Software Engineer at Qualcomm (2023-02 – Present)

Windows-on-Arm Camera

• Owned kernel-mode driver and PoFx power-management architecture for 3 camera and 1 sensing drivers (UMDF, KMDF, WDM; C/C++), covering D-, F-, and P-states; defined wake/sleep behavior and migrated drivers from WDM PoFx to WDF PoFx.
• Owned the always-on camera-sensing driver, delivering presence data to the upper presence stack in island and non-island low-power modes. For human presence, worked closely with the native SensorClx team, defining private driver interfaces, and developed the human-presence HID sensor stack (descriptors, report formats) as OEM R&D. Led newer sensing features — gesture, liveliness, posture, and face recognition — consuming and reporting inference outputs, currently shipping.
• Integrated firmware-generated on-device face-detection outputs into the Windows camera and presence pipeline, handling frame sharing while the main camera runs, preprocessing (scaling, cropping, interpolation), and direct inference consumption in always-on mode. Validated presence, distance, confidence, and face-bounding-box outputs in pre-production calibration and flagged model-accuracy issues.
• Implemented ACPI/BootFW power resources for camera drivers (boot bring-up, _DSM methods, clocks, GPIOs, I2C bring-up via Qualcomm CCI). Diagnosed and mitigated AP–ADSP messaging latency on the operation hot path across subsystems, including a wake-from-hibernation thread stall under heavy load, using latency measurements and a temporary thread-priority boost.
• Debugged BSODs, memory corruption, deadlocks, priority inversion, and performance regressions with WinDbg, KD, WPR, Driver Verifier, and AppVerifier, using live-kernel and crash-dump analysis, IRQL analysis, synchronization and reentrancy debugging, safe user-buffer access, and ETW/WPP tracing to resolve WHCP/WHQL issues.

Software Engineer Intern at Miso Robotics (2022 – 2022)

Laser-guided robotic gripper

• Developed visual servoing based gripping pipeline which improved the localization accuracy.
• Evaluated mmWave radar, ultrasonic, LiDAR, and RFID sensors alongside image sensors.
• Achieved < 1 cm localization accuracy in real time testing in a noisy, challenging environment & improved gripping accuracy from 96.5% to 98%.

Master of Science in Robotics and Autonomous Systems (Computer Science & Artificial Intelligence) – Arizona State University (2022-12)

Bachelor of Engineering in Computer Science and Engineering – Panjab University (2016-08 – 2020-06)