Rajiv Abraham Email & Phone Number

Toronto, On, Ca

Montreal, Quebec, Ca

- Led a team of two ML developers to identify and address platform debt, while strategically planning and implementing updates. - Deployed Random Forest models using MLFlow for serving on AWS Sagemaker with best engineering practices like unit testing, API replay and load testing. Minimized business risk by performing shadow deployments to validate the model before full production usage. Monitored live deployments with a combination of Datadog for service statistics and Delta Lake for debugging predictions.- Developed and optimized Airflow Directed Acyclic Graphs (DAGs) for seamless data extraction from Databricks Delta Lake using Spark and synchronization with AWS Elasticache(Redis), ensuring efficient and reliable data workflows.- Deployed jobs on the Databricks platform leveraging Spark, Delta Lake and Workflows.- Led the Proof of Concept (POC) engineering effort for A/B testing models using Statsig. Successfully identified Statsig as the vendor of choice after comparing it with another vendor in live experiments. Delivered the project on time, disseminated knowledge across the organization, and provided ongoing support to various projects with expertise in A/B testing methodologies.- Advocated for and prototyped robust monitoring systems using Databricks Lakehouse Monitoring to detect feature drift.- Conducted feature store optimization benchmarking showing the optimal compression and serialization technologies to get up to 60% space savings in AWS Elasticache(Redis). Wrote locust like load testing scripts for benchmarking latencies of different Elasticache configurations for production workloads.- Researched and presented new technologies like Featureform, OpenTelemetry, Flyte to the ML Platform team.

Toronto, Ontario, Ca

Wisedocs applies models/heuristics to medical documents to derive insights relevant for reviewing insurance claims.- Built and maintained complex model(Tensorflow, PyTorch) pipelines to extract information from pdfs and combine the predictions with heuristics.- Applied sound design abstractions enabling maintainable code to detect forms making it easy to add heuristics.- Instrumented prediction pipelines to analyze metrics for different models over time.- Deployed a multi-model COVID antigen kit image classification project using Ray Serve. A Detectron2 model identified the result window and a Resnet34 model classified the result window into a positive or negative result.- Introduced docker containerization for accurate bug reproducibility.- Wrote ETL pipelines using Ray, an actor based ML framework- Created a self-serve framework for data scientists to quickly deploy their models to production in a Ray environment. - Created an evaluation framework to assess a new model when we add them to our complex ML models pipeline.- Designed a modular architecture to quickly combine heuristics with models to enhance predictions of page attributes like dates, document titles etc.- Created an efficient error analysis workflow to quickly identify incorrect predictions filtered on different criteria to improve our heuristics.- Created and maintained a comprehensive regression suite, including property based automated testing using Python Hypothesis for exploring edge cases.- Created a feature store interface to quickly find features depending on complex filters or SQL code fragments.- Brought in business metrics for the ML pipeline using DuckDB so that we could assess our weekly progress.- Improved prediction of messy data with fuzzy matching.

Toronto, Ontario, Ca

- Introduced Determined.ai to the company to improve model training experience.- Performed load testing on prediction servers with Locust.- Created a wrapper of docker scripts for rapid experimentation and ease of developer experience, where one could use their favorite IDE locally but run the code on a remote docker container using docker contexts. Code syncing between laptop and remote machine was done using mutagen. - Updated inference pipeline for SOC2 compliance using OSV Schema, Trivy and AWS Inspector. - Mentored and provided design guidelines to junior programmers. See LinkedIn Reviews for details.- Passionate about dev experience. Creating tools/design to make it easier to debug issues or analyze data- Created ML Product and MLOps roadmaps for product owners and executives.

Toronto, Ontario, Ca

15Rock is a startup that enables business analysts to publish their analysis using a Jupyter notebook on 15Rock's serverless platform, supported by easy to create data stores.- Integrated JupyterHub into the 15Rock platform. Built the feature to take a notebook and deploy it to the Nuclio serverless platform.- Developed an extension which allows users to deploy Flask applications to the Nuclio platform.- Developed the database service in Flask which allowed users to create multiple datastores and access them conveniently. - Deployed the platform on Kubernetes on Google Cloud Platform, including load balancers and TLS encryption and connecting via Kubernetes to Google Cloud SQL.- Leveraged state of the art services like Kong API Gateway(token based authentication), Keycloak(user management), MinIO(AWS S3 replacement).

Toronto, Ontario, Ca

- Built a recommendation service prototype for content recommendation using LightFM. LightFM is a hybrid matrix factorization model specialized to solve cold starts. Wrote unit tests for feature engineering.- Utilized hyperopt for hyper parameter optimization of the model.- Built a serverless ML prediction system on AWS Lambda(github:scoremedia/thampi). By adhering to a simple Python interface, a data scientist can deploy a model without any operations knowledge with a single command.- Disseminated knowledge on machine learning and software engineering practices internally across teams.- Implemented the ETL Layer of our analytics pipeline on AWS Lambda which fed the transformed data into Amazon Kinesis Firehose which loaded the data into Amazon Redshift.- Evangelized Airflow and introduced it to the team.

Toronto, Ontario, Ca

- Built an admin UI for our authentication service- Maintained our internal authentication/authorization service using Rails- Engineering and maintaining Ruby/Rails core services integrating with Elasticsearch, Beanstalk

Toronto, Ontario, Ca

- Member of the emerging technologies team implementing the next generation of payment applications using an asynchronous variant of flow based programming(FBP) as the underlying runtime.- Developed the command line debugging and coverage tools for the FBP runtime engine.- Successfully ported the Prolog prototype of the FBP compiler to Common Lisp.- Automated the FBP compiler prototype on Windows with PowerShell scripts.

- Provided general consulting services for Java development- Developed a prototype for prediction of mutual fund performance with Spark(0.7.2) and Scala(2.9.3).- Developed code test-driven using ScalaTest.

Walldorf, Bw, De

- Coach for XP practices within SAP. This involves teaching Test Driven Development, pair programming, clean code and isolating legacy code for unit testing. This is a one week course. As a trainer, the participants rated us at 9.6 out of 10.- Organized a Coding Dojo(Practical Hands on sessions) at work for 6 months. The purpose was to introduce the department to Test Driven Development and encourage discussion. - Constant focus on applying TDD to legacy systems.Technologies worked on: Java EE, Mockito.

Montreal, Qc, Ca

- Honed my analytical and writing skills during my thesis. The thesis was accepted without any changes ( "as is" ) which is a rare honor. All thanks to the Dependable Software Research Group ! (http://users.encs.concordia.ca/~dsrg/main/people/)- Case study in applying Model Driven Development to a non-trivial project.- Solving the tricky problem of model-code synchronization with a state of the art tool

Montreal, Qc, Ca

Tutor for Course: Software Design - Enterprise Application Patterns, Design Patterns (Gang of Four) and Refactoring.- Eclipse WTP- Conducted presentations to groups of thirty

- Developed web pages in ASP.NET for an online recruitment site.- Developed stored procedures and triggers for managing databases in SQL Server 2000.- Oversaw the transition of 5 senior developers to the project during a critical phase without compromising on quality- Was the focal point for the Business Analyst and Quality Assurance Team.

Masters, Computer Science

Bachelor, Computer Engineering