Merchant Foundation Model team

• Fine-tuned a Mistral-7B LLM-based entity extraction model on unstructured data using Ray achieving 85% accuracy

• Built 2 new open source providers for Apache Airflow to orchestrate Ray jobs as a part of Airflow tasks. See below.

https://www.astronomer.io/ray/

https://github.com/astronomer/astro-provider-ray

Presented at Airflow summit 2024 on scheduled LLM fine tuning and batch inference at scale.

ML Engineer on Big Data & AI platform team, Microsoft Advertising

• Deployed AI models in production for Ads Ranking, Search Intent Understanding & Online bidding

• Optimized Spark Streaming pipelines to reduce per mini batch execution time and improve reliability

• Helped optimize storage utilization which resulted in 50% reduction in storage costs

• Worked on a project to compress our AI models using Quantization & Pruning to reduce latency by ~ 70%

✔ Innovation - Product integration b/w TIBCO Data Science and AWS Sagemaker to build distributed TensorFlow models on the cloud. Demo presented at AWS Re:Invent 2018.

* Creating new custom operator for our data science platform using Apache Spark & Scala. These operator make AI tasks like Image recognition easier at scale.

* Built an integration between Azure ML and TIBCO data science to train an autoencoder on multiple GPUs. This capability helps Read, Transform & Train DL models at scale. Used Python, TensorFlow, Azure ML, Horovod etc.

* Trained Image recognition models using TF's dataset API and CNNs on multiple CPUs. Used pyspark, AWS Sagemaker & Distributed Tensorflow

✔ Customers - Closed $5M+ value deals with customers who needed to build Neural nets on Big data

✔ Community - Built a demo for anomaly detection using Python, Tensorflow etc.

✔ Innovation - EdgeML (Inferring tensorflow models on IoT devices), Product Integrations

✔ Customers - Build POC demos for high value deals and presented to senior management

✔ Community - Built ML templates to enable citizen data science - XGBoost, Random Forest, Autoencoders etc. Also, contributed tensorflow models to open source projects