* Build foundation bilingual language models, for English and Arabic domain using Scaling laws of LLMs.* Experiment on different architectural choices in areas like long context extension with RoPE and Alibi, Attention choices between Multi Head, Group Query and Sliding Window, muP transfers for initialisation and learning rate. * Experiments on Mixture of Experts style of model with Branch Train Mix techniques.

* Leads two engineering teams (Machine Learning and NLP) and drive product roadmap for GenerativeAI solutions in Education.• Work closely with C-level executives and Directors for product strategy, capacity planning, and hiringengineering resources, managing cloud operations in AWS and Azure, infra allocation, and budgetingplans.• Developed a contextualized tutor for providing step-by-step hints to solve math problems using sentencetransformers + langchain + GPT4 + faiss… Show more * Leads two engineering teams (Machine Learning and NLP) and drive product roadmap for GenerativeAI solutions in Education.• Work closely with C-level executives and Directors for product strategy, capacity planning, and hiringengineering resources, managing cloud operations in AWS and Azure, infra allocation, and budgetingplans.• Developed a contextualized tutor for providing step-by-step hints to solve math problems using sentencetransformers + langchain + GPT4 + faiss vector store.• Built a QnA system for generating assessment questions from a given context and data sources llama-2+ llama-index + langchain in AWS using Bedrock.• Fine tune foundational models like llama-2, Falcon, Mistral etc with instructions using PEFT-LoRA oneducational datasets and quantize models for inference optimization.• Extract and process text and image data sources using llama-index and build a (retrival augmentedgeneration) RAG pipeline using vector databases for storing and retrieving embeddings Show less

* Lead a team of machine learning engineers for developing personalized recommendations and NLPapplications.• Built a quiz application for 100k daily active users (DAU) scaling with various pipeline optimizationsdownstream finetuning using RoBERTa and BART-based models.• Developed a lesson plan and assessment recommendation system using Collaborative Filtering thatpredicts the next sequence of lessons and questions for students achieving 13% learning improvement.•… Show more * Lead a team of machine learning engineers for developing personalized recommendations and NLPapplications.• Built a quiz application for 100k daily active users (DAU) scaling with various pipeline optimizationsdownstream finetuning using RoBERTa and BART-based models.• Developed a lesson plan and assessment recommendation system using Collaborative Filtering thatpredicts the next sequence of lessons and questions for students achieving 13% learning improvement.• Predicted top 20 influencing factors for student's learning from TIMSS 2019 assessments using Catboost+ SHAP interpretations.• Developed ML infra services for data processing, training, and inference using AWS EKS, ECR, S3,MongoDB, kubeflow and mlflow. Show less

* Predicted scoring ability of students in exams using XGBoost with 78% accuracy and RMSE less than8.7.• Developed a system for predicting student grade gap and churn estimation using LSTM network.• Built and scaled streaming and batch pipeline of 500k users using spark streaming, kafka, ksql, S3 andRedshift.* Building a "structured streaming" pipeline for real time processing, rule based alerting and applying machine learning algorithms and learning the challenges… Show more * Predicted scoring ability of students in exams using XGBoost with 78% accuracy and RMSE less than8.7.• Developed a system for predicting student grade gap and churn estimation using LSTM network.• Built and scaled streaming and batch pipeline of 500k users using spark streaming, kafka, ksql, S3 andRedshift.* Building a "structured streaming" pipeline for real time processing, rule based alerting and applying machine learning algorithms and learning the challenges around reactive and distributed systems. Stack: Apache Spark, Structured Streaming, Spark MLLibs, Apache Kafka, Scala, Amazon AWS, S3, RedShift, Docker, MongoDB, Tableau, Elastic Search, Kibana Show less

* Developed a scoring for risk anomaly model in high-value transactions (HVT) using gradient boostingtechniques.• Built a customer segmentation model using PCA for dimensionality reduction and KNN for groupingthem into segments.• Achieved 80% cost savings by improving the latency of the analytics system from 45 to 250 TPS for ascale of 6 billion events per day• Designed lambda architecture by consuming real-time events from Kafka for data analytics.•. Stack: Scala… Show more * Developed a scoring for risk anomaly model in high-value transactions (HVT) using gradient boostingtechniques.• Built a customer segmentation model using PCA for dimensionality reduction and KNN for groupingthem into segments.• Achieved 80% cost savings by improving the latency of the analytics system from 45 to 250 TPS for ascale of 6 billion events per day• Designed lambda architecture by consuming real-time events from Kafka for data analytics.•. Stack: Scala, Akka-Actors, Akka-Streams, Kafka, Hadoop, Squbs, Akka-HTTP, Json4sOpen Source & Tech Talks• Problem: To avoid data loss in a distributed cluster of machines with local persistence.Contribution: A backpressure-aware pub-sub buffer using a chronicle queue capable of publishing data to multiple subscribers, each at its own pace managed by a commit log.Speaker at Great Indian Developer Summit, April 2017, Bangalore, India. Topic: “Streaming Behavioral Analytics & Building a Scalable Data Pipeline” Show less

Zoho Invoice, Zoho Expense and Zoho BooksIntegration Engineer – Payments• Designed a generic framework handling online payments lifecycle - authorize, capture, payment, dispute & settlement with multiple payment gateways• Implemented in-app purchase for mobile clients and recurring invoice transactions.• Stack: Java, Spring, Dependency Injection, Redis, Payment SDKsApplication Developer – Finance Apps (Invoice and Accounts)• Developed RESTful APIs for… Show more Zoho Invoice, Zoho Expense and Zoho BooksIntegration Engineer – Payments• Designed a generic framework handling online payments lifecycle - authorize, capture, payment, dispute & settlement with multiple payment gateways• Implemented in-app purchase for mobile clients and recurring invoice transactions.• Stack: Java, Spring, Dependency Injection, Redis, Payment SDKsApplication Developer – Finance Apps (Invoice and Accounts)• Developed RESTful APIs for transaction based application handling financial data.• Optimized complex SQL queries using explain, group-by & joins and got < 300ms response time.• Stack: Java, MySQL, JAXB, Tomcat, JSON (Jackson parsers) Show less

Zoho is cloud company developing a suite of business, productivity, finance, collaboration, sales, marketing and helpdesk apps.Key Responsibilities:* Started working on code quality, corner case identification and code coverage in Java.* Learned software build environment management with Apache Ant, Ruby Rake & Gradle.* Exposure to version control system like Mercurial (Hg) and Git.* Server life-cycle management with Apache Tomcat. Java server technology like… Show more Zoho is cloud company developing a suite of business, productivity, finance, collaboration, sales, marketing and helpdesk apps.Key Responsibilities:* Started working on code quality, corner case identification and code coverage in Java.* Learned software build environment management with Apache Ant, Ruby Rake & Gradle.* Exposure to version control system like Mercurial (Hg) and Git.* Server life-cycle management with Apache Tomcat. Java server technology like Servlets and JSP.* Exposure the JDBC APIs and database connections in MySQL* Covered unit test cases & automated test report generation with Junit & Cucumber.* Wrote Shell (bash - Mac OSX) & Ruby scripts to avoid manual task and improved automation. Show less