Exploring Weak-Strong Model Dynamics for Robustness Against Dataset Artifacts in MultiNLI
Emmanuel Rajapandian
[Code]
Cover Letter
I work towards enhancing ASIN search impressions & conversion rates for Amazon US Stores by building ETL data pipelines and pooling the data through ML, Generative AI tech stack such as LLMs, RAGs to extract insights & build analytical solutions. I have over 4+ years of dedicated experience in ML/AI, including roles as a Data Scientist and Business Intelligence Engineer/Analyst. Since mid 2023, I have been interested in Causal ML trying to evaluate the cause and effect of business usecases with SOTA models. I have also been exploring LLMs, using advanced prompting technqiues, guardrails to address and help solve customer pain points in WW Amazon Stores. I love to solve problems which improve the customer experience and cherish new and complex challenges.
Education
- M.S. in Datascience, University of Texas, Austin, USA
- Completed Coursework: Data Exploration & Visualization (A), Probability and Simulation Based Inference (A), Data Structures & Algorithms (B+), Foundations of Regression and Predictive Modeling (A), Advanced Predictive Models (A), Design Principles & Causal Inference (B+), Deep Learning (A), Convex Optimization (B+), Machine Learning (B+), Natural Langugage Processing (A)
- B.Tech. in Electronics and Computer Engineering, Indian Institute of Information Technology, Chennai, India
- Relevant Coursework (“A” grades): Data Structures and Algorithms, Digital Image/Signal Processing, Data Analytics in Python, Probability Theory, Information Theory and Coding.
Work experience
- Amazon - Business Intelligence Engineer II - [Apr 2025 -]
- Currently working on launching a self-service analytical platform integrating complex data sources along with natural language capabilities for any user to access data-driven insights.
- Amazon - Business Intelligence Engineer I - [Oct 2022 - Mar 2025]
- Individual contributor with responsibilities such as formulating the business opportunities, extract data, solving them using ML/NLP to enhance customer buying experience.
- Promoted within 2.5 years for consistent delivery of high impactful solutions in multiple programs.
- Tech Stack: AWS Sagemaker, Lambda, S3, EC2, Clouwatch, Redshift, Bedrock.
- Applied Data Finance - Jr. Data Scientist; Applied R&D - [May 2021 - Sep 2022]
- Spearheaded data-driven initiatives for a US Loan Portfolio Channel by architecting and productionizing ML-based credit UW models, driving a 20% reduction in loan delinquency through behavioral risk segmentation.
- Built XGBoost & LightGBM pipelines to detect defaulters in risky loan channels, achieving 0.55 on Kolmogorov–Smirnov scale.
- Ensured lending regulatory compliance by performing feature engineering, model building, & explainability.
- Co-owned a pivotal analytics project, shaping outcome decisions through leadership by actively engaging with senior management, chief analytics officer, to strategise and drive initiatives.
- Drove 800K dollars in monthly profits and sustained 15% Cost Per Funded Dollar through CI/CD enabled model optimization.
- Additionally, increased profits by 15% by implementing Bayesian optimization to tune Experian & TransUnion parameters.
- Architected a Markov chain simulation framework of the application funnel — improving loan conversion by 5%.
- Tech/Skill Stack: ML, PostgreSQL, Business/Predictive analytics, Risk modelling.
- Larsen & Toubro Sciences R&D - Datascience Intern - [Jul 2020 - Oct 2020]
- Leveraged Elasticsearch to elevate semantic search functionality within L&T’s internal website.
- Fine-tuned and seamlessly integrated a search next-word predictor using the then newly released Google T5 model within the L&T database.
- Tech/Skill Stack: NLP, ML, Python.
Skills
- ML Techniques: Regression, Classification, Boosted ML, Recommendation, Ranking, Causal ML.
- Frameworks: PyTorch, Scikit-Learn, TensorFlow, HuggingFace Pipelines, Transformer.
- Cloud/Tools: AWS (Sagemaker, Lambda, Glue, Athena, S3, Cloudwatch, Redshift, Bedrock), QuickSight, DataGrip.
- LLMS & AI: Mistral, GPT-3.5, Claude, T5, BERT, LangChain, RAG, vLLM.
- Languages: Python, PostgreSQL, SQL, Spark SQL, Scala, R.
- Soft Skills: Ownership, Leadership, Agile Mindset, Adaptability, Communication, Business Document Writing.
Projects
Transformer Language Modeling
Emmanuel Rajapandian
[Code]
Feedforward Neural Networks, Word Embeddings and Generalization
Emmanuel Rajapandian
[Code]
Linear Sentiment Classification
Emmanuel Rajapandian
[Code]
Autonomous RL-DL Agent for Realtime multiplayer SuperTuxKart Ice-Hockey
Emmanuel Rajapandian, Jean Del Rosario Pegeuro
Image based vision system for a racing simulator - SuperTuxKart
Emmanuel Rajapandian
RAG Chatbot using Falcon 7B LLM using Langchain
Emmanuel Rajapandian
[Code]
Other roles and leadership
- Society of Automotive Engineers Club (IIIT) - Team Captain - (Aug 2019 - Mar 2020)
- Captained the college SAE team comprising of 7 members, tasked to design and model a RC plane.
- Successfully modeled and flight-tested 10 prototypes within a span of 4 months.
- Chennai Cubing Club (c^3) - Core Organizer - (Aug 2018 - )
- Successfully organized and conducted IIITDM Cube Open in 2018.
- Core member in Chennai Cubing Club for conducting WCA competitions across Tamilnadu.
- Currently ranked top 10 in India in Pyraminx category. Was ranked top 30 in the world (2018-2020) and was the runner-up national champion in this event.