David Wright

Senior Software Engineer | ML Inference, Classification, Search & Extraction | Distributed Systems | C++

ML Pipelines/Search & Retrieval: Owned document scanning & classification, descriptor/embedding-based search and retrieval, and ML inference. Designed in-memory indexes with query APIs.
Data Infrastructure: High-throughput streaming and data pipelines (up to 1GB/s), distributed platforms serving 10M+ requests/year across SaaS, on-premises, and mobile.
High-Performance Software: 10+ years of C++ and Python across real-time sensor pipelines, distributed services, and embedded systems, with production experience in Rust, Kotlin, and Java.

Professional Experience

Co-owned distributed high-bandwidth (~1GB/s) streaming and recording solutions for sensors, cameras, and device state (SLAM), using gstreamer.
Owned sensor metrics validation tool, defining new metrics and performing online statistical validation (t-digest) of streaming metadata (drops, jitter, synchronization) across XR devices.
Led a platformisation initiative for sensor and camera validation, resulting in a 95% reduction of developer effort for onboarding new XR devices.
Lead developer on always-on data collection tool for XR devices. Designed mmio/mmap buffering strategy achieving 300MB/s sustained writes without noticeable user impact.

Owned the ML inference infrastructure and backend services powering the CV/ML stack, serving 10+ million verification requests per year.
Led a team of 3 engineers owning the platform of the company’s distributed Identity Verification system across SaaS, on-premises, and mobile deployments.
Designed and led passive injection attack detection using a dual-pipeline architecture (UX-optimized vs. security-optimized) to defend against video replay attacks while maintaining conversion.
Led upgrades to video frame extraction (quality-based algorithm, +60% data yield) and snippet extraction (text-boundary detection, +20% OCR).

Tuned field detection algorithm using Optuna hyperparameter optimisation, achieving up to 10% improvement.

Designed the search and retrieval pipeline for an ML document embedding model, defining cosine-distance nearest-neighbour search with ratio-test rejection against a reference database to achieve >90% recall.
Designed gRPC and WebAssembly SDKs that moved document detection and quality checks client-side, cutting per-session backend calls by ~50%. Adopted company-wide and by external customers.
Implemented a nonlinear optimization backend for document unwarping, improving data extraction quality.

Designed an in-memory multi-source index and its query API, unifying document fields extracted across multiple pipelines (OCR, document classifier) for fast downstream consumption.
Designed and implemented a Gabor filter-based image descriptor for document classification and retrieval (~85% recall).

Developed ML-based and classical CV algorithms for data extraction and security features, deploying models via ONNX Runtime for cross-platform inference.
Benchmarked and integrated face-verification and liveness detection models in the identity verification pipeline.

Reduced real-time pipeline latency 30x (from ~300ms to <10ms) through profiling (gperftools, perf, Apple Instruments), parallel programming (OpenCL, SIMD), and mathematical optimisation.
Designed neuromorphic monocular depth estimation algorithms for event-based processing, achieving <1mm accuracy at <10ms latency and 5kHz sampling rate.

Implemented data clustering and segmentation algorithms for neuromorphic sensor calibration.

Developed SIL4 train protection software for Siemens Trainguard ETCS on-board unit, a platform deployed on 6,000+ trains across 50+ countries.
Designed and evaluated slip/slide detection algorithms using Extended Kalman Filters, wavelet analysis, statistical tests, and SVMs.

Developed SIL3 functional safety firmware for industrial drive supervision, deployed on 100+ drives in production.
Implemented multiprocessor synchronization for drive position and velocity supervision.
Designed PCBs for I/O testing of drive hardware.


MSc. Engineering (part-time)	HSR University of Applied Sciences Rapperswil	2017-2021
BSc. Electrical Engineering	HSR University of Applied Sciences Rapperswil	2011-2014
Apprenticeship	Bischofberger AG Zürich: Electrician	2006-2010
Vocational College	Gewerbliche Berufsschule Wetzikon	2006-2010
Vocational Baccalaureate	Berufsmaturitätsschule Zürich	2006-2010

Type	Title	Grade
Master Thesis	Neuromorphic Vision Based 3D Laser Mapping and Robot Control	6.0 / 6.0
Project Thesis	A Reinforcement Learning Based Rover Navigation	5.5 / 6.0
Bachelor Thesis	Piezoelectric Energy Harvesting For HSRVote	6.0 / 6.0
Student Research Thesis	Optimal Sensor Calibration using Lasso Regression	5.5 / 6.0

Appreciation Award issued by Electrosuisse AG for the Bachelor Thesis Piezoelectric Energy Harvesting for HSRvote

Swiss German (native), German (bilingual), English (bilingual), French (elementary)