Aamir Khan Email and Phone Number

• Worked on terabytes of insurance data related to claims, policy, third-party economic data (David-Gordon, TransUnion, etc.), geographic/demographic data, and peril-specific information stored in traditional databases (Oracle, Hue/Impala, SAS, Share point) and Google BigQuery.• Q3-Q4 2022: Part of the team to implement and deploy a claims fraud flagging model specific to property fire peril. Flags top 20% suspicious claims with 24% precision. Used gradient boosting model with log-loss metric. Net annual benefits of the solution are estimated around $450k.• Q1 2023: Refreshed model used to deflect suspicious policy applications in Sonnet online insurance tool. Worked on H2O (Random Forest, stacked ensemble) with AUCPR as a performance metric. As a customer-facing high-risk model, performed a detailed bias-fairness analysis comprising six demographic details.• Q2 2023: Part of the team implementing property theft by-peril fraud detection model. Provides a net new annual benefit of over $315k compared to prior manual triage.• Q2-Q3 2023: Worked on many PoC business rules for Wind/Water/Hail related claims fraud detection of which 4 rules were finally implemented.• Q2-Q3 2023: Worked on various kinds of analysis and benefits estimation reports for newly introduced/decommissioned/refreshed models as well as business rules.• Q2-Q4 2023: Provided leadership role for the initiative to migrate, establish, and organize model repositories on Collibra to attain enhanced model risk management for all fraud related models.• Q3-Q4 2023: Worked on the massive migration of processes to the Google Cloud Platform with modifying key modules to ingest data from BigQuery, reimplementation of models in Vertex AI, and deploying end-to-end ML pipeline projects.• Q4 2023 – Q1 2024: Worked on an MVP of Generative AI, LLMs (PaLM 2 Text Bison, Llama 2), NLP techniques (BERT variants, embeddings) to extract significant value from property claim notes provided by adjusters over time.

Every winter semester, teaching two 1.5 credit Master's level courses:The ‘Cloud Computing’ course focuses on the fundamentals of cloud computing focusing on basic services like Compute, Storage, various IAM settings, Networking and VPC configurations, Databases (Cloud SQL and BigTable), and BigQuery data warehouse.The ‘Data Analytics, Machine Learning, and AI on Cloud Platforms’ course covers advanced topics in creating Dataflow pipelines, Pub/Sub, and creating ML models in Vertex AI notebooks. Machine learning on Vertex AI includes importing custom datasets, creating predictive models, deploying to endpoints, and incorporating project work into pipelines. Lastly, it covers the case study of working on various BigQuery ML models.https://experts.mcmaster.ca/display/khana513

Data specialist working with logs and telemetry data from a diverse set of devices installed in hundreds of communication tower sites across Canada (Bell & Rogers) and Cambodia. Regularly collected data from various Bell and Rogers mobile communication tower sites in Mississauga. Worked through all phases, from data cleansing, integration, and feature selection to modeling and delivering final predictive models. Worked with Google Cloud services (BigQuery, Data Studio, etc.), InfluxDB, time-series analysis, anomaly detection, machine learning libraries, Python, PySpark, Scikit-learn, Pandas, Grafana, and Matplotlib.• Project (Fuel forecasting using time-series ML model) – The project was about analyzing historical telemetry data from Cambodia sites to build a model that can forecast fuel consumption for the next month. Over 500GB of raw telemetry data was ingested and cleansed for further processing. Two main sources of anomalous data were identified: faulty sensors and staff malpractices. Time-series model libraries used were Darts & NBeats. Actions taken resulted in 20% fuel efficiency at selected sites. • Project (Energy profiling & anomaly detection of Bell site) – The project was to identify patterns in power production from wind, solar, and diesel generator. Data was mixed with third-party weather data to enhance understanding. Anomalies were reported in solar power production which was due to blocked solar panels. The proposed protocol resulted in a 15% increase in solar power production.• Project (Data integrity verification of AQue Global system) – Created Python scripts that periodically validate data ingested into the AQue Global (a proprietary system, product by PLC Group https://www.plcgroup.com/) database from a diverse set of sources. This practice helped us identify a very common bug in InfluxDB 1.8 implementation regarding column datatypes saving a ton of time that the business was investing in manual blending and monitoring of data.

• UBC MDS program capstone project.• Implemented IT tickets triage model using state-of-the-art NLP and techniques. IT ticket requests were automatically classified into three categories: Impact (levels 1,2, or 3), Severity (levels Low, Medium, or High), and Board (target service department: Sales, Incidents, or Service Requests).• Worked with word embedding methods like BERT/TF-IDF, models like SVM, Naive Bayes, XGBoost, and transformer models like BERT/RoBERTa. Deployed models using Docker and Flask• Reduced human efforts worth two persons each day by automating the whole process.

Taught computer science and data analytics skills to undergraduate and graduate students for over sixyears. Specialized in Operations Research, OOP, Algorithms, Data Structures, Databases, Artificial Intelligence, and Machine Learning. Some course lectures are available on my YouTube channel: https://www.youtube.com/user/jadoonengr79/playlists