Nov 2024 – Present · 1 yr 4 mos · San Francisco, California, United States

• Helping businesses unlock new markets with localization

Feb 2024 – Oct 2024 · 8 mos · London, England, United Kingdom · Hybrid

• Develop and execute a comprehensive machine learning strategy aligned with XYZ
• Reality's business objectives, ensuring XYZ's MR platform remains at the forefront of
• technological advancements.
• Build, mentor, and manage a high-performing team of machine learning engineers
• and data scientists, fostering a culture of innovation, collaboration, and continuous
• learning.
• Lead research initiatives to identify new opportunities and applications for machine
• learning in the construction industry, exploring areas such as computer vision, natural
• language processing, predictive analytics, and more.
• Drive the development of cutting-edge machine learning algorithms and models to
• enhance the capabilities of XYZ's MR platform, including object recognition, spatial
• mapping, and 3D visualization.
• Collaborate with cross-functional teams to deploy machine learning models into the
• MR platform, ensuring scalability, reliability, and real-time performance.
• Continuously monitor and improve the performance of machine learning models,
• identifying and addressing issues to enhance user experience and efficiency.

Oct 2022 – Feb 2024 · 1 yr 4 mos · London Area, United Kingdom · Remote

• Fight Fraud with Machine Learning teaches you to build and deploy state-of-the-art fraud detection systems. You’ll start with the basics of rule-based systems, iterating chapter-by-chapter until you’re creating tools to stop the most sophisticated modern attacks. Almost every online fraud you might encounter is covered in detail.
• Examples and exercises help you practice identifying credit card fraud with logistic regression, using decision trees and random forests to identify fraudulent online transactions, and detecting fake insurance claims through gradient boosted trees. You’ll deploy neural networks to tackle Know Your Customer fraud, spot social network bots, catch deepfakes, and more! Plus, you’ll even dive into the latest research papers to discover powerful deep learning techniques such as vision transformers.

Sep 2022 – Feb 2024 · 1 yr 5 mos · London Area, United Kingdom · Remote

• PyTorch is making it easier than ever before for anyone to build deep learning applications. This PyTorch book will help you uncover expert techniques to get the most from your data and build complex neural network models.
• You’ll create convolutional neural networks (CNNs) for image classification and recurrent neural networks (RNNs) and transformers for sentiment analysis. As you advance, you'll apply deep learning across different domains, such as music, text, and image generation using generative models. You'll not only build and train your own deep reinforcement learning models in PyTorch but also deploy PyTorch models to production, including mobiles and embedded devices. Finally, you’ll discover the PyTorch ecosystem and its rich set of libraries. These libraries will add another set of tools to your deep learning toolbelt, teaching you how to use [fast.ai](http://fast.ai/) for prototyping models to training models using PyTorch Lightning. You’ll discover libraries for AutoML and explainable AI, create recommendation systems using TorchRec, and build language and vision transformers with Hugging Face.
• By the end of this PyTorch book, you'll be able to perform complex deep learning tasks using PyTorch to build smart artificial intelligence models.

May 2020 – Feb 2021 · 9 mos · London, England, United Kingdom

• Authored first edition of "Mastering PyTorch" - Build powerful neural network architectures using advanced PyTorch 1.x features.
• Amazon US: https://www.amazon.com/gp/mpc/ACU0AD60YWJEU
• Amazon UK: https://www.amazon.co.uk/Mastering-PyTorch-architectures-advanced-features/dp/1789614384
• Amazon India: https://www.amazon.in/Mastering-PyTorch-architectures-advanced-features/dp/1789614384
• Packt website: https://www.packtpub.com/product/mastering-pytorch/9781789614381

Oct 2019 – May 2020 · 7 mos · London, England, United Kingdom

• Co-authored second edition of The Supervised Learning Workshop.
• Amazon: https://www.amazon.com/Supervised-Learning-Workshop-Interactive-Understanding/dp/1800209045
• Packt: https://www.packtpub.com/product/the-supervised-learning-workshop-second-edition/9781800209046

Aug 2020 – Present · 5 yrs 7 mos · London, England, United Kingdom

• Helping SUIND build and embed deep learning based models for their drone autonomy systems.
• Fine-tuned segmentation models (DeepLabV3Plus) towards tasks such as crop segmentation, agricultural field instance segmentation, and so on.
• Trained object detection models (ResNet) such as landing marker detection to automate drone landing
• Converted Torch models to C++ friendly format using Torchscript, and deployed models on Jetson Nano for inference
• Coaching AI/ML folks to execute different ML projects, rehashing existing work and compounding model performances using experiences across projects.
• Helping the CTO with cloud/infra decisions, tech hiring, brainstorming on high-level tech strategy.

Dec 2019 – Mar 2024 · 4 yrs 3 mos · London, England, United Kingdom · On-site

• Architecting in-house Computer Vision product for KYC and fraud detection; onboards 40k+ Revolut clients per day.
• Fine-tuning Swin and BART Transformers (DONUT) to auto-extract information from identity documents.
• Deployed first heavyweight (0.5 bn+ params) ML models on GPUs at Revolut using Kubernetes on GCP with FastAPI at <1s 99p latency.
• Developed a FSL (Few-Shots Learning) framework (SiameseNet + KNN) to catch fraud attacks coming from fraud rings.
• Mobile and web based document segmentation models using tflite and tfjs at <30ms 99p latency.
• Wrote AirFlow/SQL ETLs, built Looker, Metabase dashboards to track block rate, precision, FPR of ML models.
• Designed data labelling tool specs including backend (API calls, DB ops) and frontend (UI) work.
• Built tabular models using CatBoost to detect identity fraud based with >99% precision and >80% recall, on user metadata available during onboarding.
• Managing, coaching/mentoring data scientists and python engineers in the team, establishing a collaborative team culture.

Oct 2018 – Sep 2019 · 11 mos · London, United Kingdom

• 1. Building Car Insurance Claims Fraud Detection AI Pipeline.
• Claims data is tabular but we also use visual classifier features from damaged car images.
• A Multi Task DNN (PyTorch) model performs fraud detection, saving money for the insurer, and time for the insured.
• 2. Developed Language Model based Text Classifier using Fast-AI's text module to map insurance claim description lines to car parts.

Jun 2017 – May 2018 · 11 mos · Belgium

• Working with the TFF Organization towards the aim of feeding 9+ billion people by 2050.
• Concretely, I am responsible for communications and inductions within the TFF Belgium Chapter, which involves spreading the word around, encouraging (willing-to-be or already) food entrepreneurs to participate in the TFF competition and subsequently join us at the Annual TFF Academy & Summit.
• More info here : https://www.tffchallenge.com/

May 2016 – Aug 2018 · 2 yrs 3 mos · Antwerp Area, Belgium

• Leading the data science team dedicated towards and driven by mobile sensor data (accelerometer, gyrosope, GPS)
• Projects:
• 1. Timeline event prediction using LSTM network
• 2. Smarthome solutions - automatic room-type detection, sleeping patterns, etc.
• 3. Transport mode classification service using GPS-only data from mobile devices.
• 4. Trajectory embeddings microservice using per-waypoint transport-mode probabilities.
• 5. Command line tool for data scientists to manage aws instances, volumes, etc.
• 6. Dashboard/visualization tool for the data science team.

Aug 2015 – Jan 2016 · 5 mos · Stuttgart Area, Germany

• Audio Events Detection using Deep Neural Networks.
• 5-seconds audio clips transformed into MFCC features.
• With RF model as benchmark, first trained a FC DNN.
• Then significantly better results using 1-D Conv Net with max-pooling.

Jun 2013 – Aug 2014 · 1 yr 2 mos · Bangalore

• Application developer for the PAPI (Public Application Programming Interface) team.
• Develop and test APIs for the RightNow-CX cloud services, a product of Oracle.
• Platforms - C++, PHP and XML (SOAP).