About the Role
Customer addresses, Geospatial information, and Road-network play a crucial role in Amazon Logistics' Delivery Planning systems. The Geospatial science team owns exciting learning problems in the areas of Address Normalization, Geocode learning, Travel time estimations, Maps conflation, and attributes learning. These are foundational inputs that go into route planning and help in improving the delivery experience for both transporters and end customers. This role specifically focuses on problems related to address normalisation and geocode learning.
As a Senior Applied Scientist, you will be responsible for leading a team of scientists experienced in building ML models to solve complex business problems and testing them in a production environment. The scope of the role includes defining the charter for the team and proposing solutions that align with the organization's priorities and production constraints while still creating impact. You will achieve this by leveraging strong leadership and communication skills, data science skills, and by acquiring domain knowledge pertaining to delivery planning systems. You will provide ML thought leadership to technical and business leaders, and possess the ability to think strategically about business, product, and technical challenges. You will also be expected to contribute to the science community by participating in science reviews and publishing in internal or external ML conferences.
Here is a glimpse of the problems that this team deals with on a regular basis:
- Organizing addresses into a hierarchy in the presence of noisy, inconsistent, localized, and multi-lingual user inputs. This is done at the scale of millions of customers for existing as well as emerging geographies, such as North America, Europe, India, and the Middle East. We make use of technologies like entity matching, multi-modal learning, named entity recognition, and large scale predictive modelling.
- Building foundational models and knowledge graphs for Geospatial and Logistics applications. These models power multiple downstream applications such as learning delivery locations, detecting anomalies, and propagating attributes (delivery hours, holidays, and shared delivery places). The solutions require combining multiple modalities (maps, delivery locations, defects) and exploiting semantic as well as structural properties of places through the use of state-of-the-art technologies (Large Language Models and Graph Nets).
- Building generic and scalable solutions to reduce maintenance cost and enable faster expansion to many geographies. This is achieved through technologies such as zero-shot generalisation, semi or weak-supervised learning, and large scale information retrieval.
Key Job Responsibilities
- Lead one or two large scale high-impact projects.
- Implement SOTA and complex ML solutions to meet the organization's goals.
- Review and provide feedback to reporting scientists.
- Publish work in internal and external top-tier conferences.
- Provide inputs in team growth and goal planning.
- Educate peers/team on science best practices.
- Expand the use of ML within the organization.
Basic Qualifications
- 5+ years of experience building machine learning models for business application.
- PhD, or Master's degree and 6+ years of applied research experience.
- Experience programming in Java, C++, Python or related language.
- Experience with neural deep learning methods and machine learning.
- Experience in building speech recognition, machine translation and natural language processing systems (e.g., commercial speech products or government speech projects).
Preferred Qualifications
- Experience with modeling tools such as R, scikit-learn, Spark MLLib, MxNet, Tensorflow, numpy, scipy etc.
- Experience with large scale distributed systems such as Hadoop, Spark etc.
- Experience with generative deep learning models applicable to the creation of synthetic humans like CNNs, GANs, VAEs and NF.