• Working on one of complex and mature search tech stacks at Meta to improve search relevance for Facebook entity search – i.e. people, groups, creator and events search.

• Designed and developed code review time testing to reduce SEV across entire

search org

• Incorporated LLM based and Human based evaluation pipeline for better search system evaluation

• Reduced the GPU/CPU usage by trimming sub-requests based on NLP intent

• Working with product and cross-functional team to drive the Org success of search relevance improvement

• Building the latency sensitive, highly scalable and distributed backend recommendation engine and infrastructure. This engine is used for surfacing recommendations on Amazon search, homepage and products’ details page

• Design, build, test and maintain scalable and reliable infrastructure that supports machine learning workflows from training to serving ML models

• Researching and prototyping LLM based tech solution to improve customer experience and engagement

• Spending time in products’ infrastructure, operations, testing, and maintenance improvement. Working closely with different stakeholders (product managers, TPM, UX, scientists, cross engineering teams) and customers.

• Collaborate with several teams and organizations to drive the development and adoption of solutions

• Actively participating in the team's process improvement, company's hiring process, and mentoring fellow engineers

I'm one of the early developers of three AWS services.

Aws DeepComposer (AI powered music generation service) – developed training and inference ML pipelines to integrate generative models (transformer-based architecture) for music generation

https://aws.amazon.com/deepcomposer/

Aws BugBust (AI powered bug scan tool with competition feature) – designed & developed multi-tenant real time leaderboard service to manage participants’ score, ranking, rewards and prizes. Recorded Guinness world record by hosting largest bug bash competition.

https://aws.amazon.com/bugbust/

Aws DeepRacer Student (service to teach students AI & ML concepts) – developed a course tracking and quiz management service to track students' progress and scholarship.

https://aws.amazon.com/deepracer/student/

• Worked at team that builds next generation Chat-bot to automate mobile insurance claim process

• Setup backend serverless stack using Lambda, Api Gateway to serve customer facing chat bot.

• Developed end-to-end deep learning pipeline by incorporating data gathering & cleaning, preparing models and deploying on production using AWS and AzureML for user intent classification, named entity recognition, and sentiment analysis.

• Created NLP models using regular techniques like rule based models (regex), BOW (tf-idf, n-gram, LDA); state-of-arts techniques word2vec, sent2Vec, doc2vec & Glove word embedding; CNN, RNN (LSTM & GRU) based architecture for user intent classification, named entity recognition, sentiment analysis and neural language model.

• Experimented with advanced approaches like position aware self-attention, BERT etc. for intent classification

• Performed feature engineering, named entity recognition (NER) and clustering to update existing ML model.

• Created data visualization (ggplot) and dashboard (shiny) to provide insight to business problem.