United States

AI/ML Engineering

Data Scientist

AI Design

Bachelor of Engineering (B.E.)

Bijo Thomas

Sr. Applied Scientist

Problem-solving

Time management

Critical-thinking

Empathy

Stakeholder management

AI/ML

Business intelligence

Conversational AI

Data mining

Machine learning

Pandas

Python

PyTorch

Apache Spark

Jupyter Notebook

Numpy

As a Sr. Applied Scientist specializing in catalog integrity and customer service, I develop scalable machine learning solutions that enable abuse detection, automate corrective actions and optimize operational efficiency. My work has helped protect amazon customers and sellers from fraudulent content. Some highlights include:

• Designed and implemented siamese networks leveraging multilingual text embeddings to identify inconsistencies in unstructured product data and improved automated correction
• Fine-tuned multimodal LLMs reducing duplicate listings by 34% YoY and increasing automated corrections by 45%. Experimented with different fine-tuning strategies (LoRA, full fine-tuning, multi-stage, etc.)
• Prepared synthetic features and labels addressing data scarce model training using solutions ranging from classical ML to LLMs
• Led consolidation efforts across models, reducing redundancies and achieving 7x cost reduction while improving detection precision. Deployed distilled/pruned models and smart data ingestion to replace legacy pipelines.
• Used 0 shot and supervised multimodal and multilingual embeddings to detect temporal and marketplace level product inconsistencies
• Improved LLM comprehension of tabular datasets using feature statistics in a few shot setting, adding explainability and enhancing classification performance over traditional tree-based ML models
• Developed deep learning NLU models to predict coaching opportunities for customer service associates by fine tuning pre-trained language models using annotated samples. Predicted coaching themes, improved coaching efficacy, handle time and transfer rates
• Collaborated with product managers, risk managers, scientists, and engineers to define problems, review designs, build frameworks and implement scalable ML solutions
• Published work in Amazon internal ML Conferences

• Developed deep learning models using Customer Service (CS) contact transcripts to recommend coaching-relevant contacts, improving CS associate training efficiency; built the science infrastructure to train, evaluate, scale and productionize these models
• Designed and executed A/B testing on 10+ customer experience survey variations, optimizing UI and question formats to improve response rates and minimize customer effort
• Built contact difficulty prediction model enabling intelligent contact routing to tenured associates
• Established an annotation program using AWS Ground Truth, automating high-quality labeled data generation to train ML models
• Designed experiments, conducted code reviews, scaled deployments across verticals and improved ML models iteratively
• Developed automated data pipelines and dashboards, ensuring robust experiment tracking.
• Mentored engineers and scientists, expanding data science capabilities within the team.

• Analyzed customer satisfaction surveys using synthetic cohort analysis to measure the impact of customer experience on spending. Insights drove key investment decisions (ex: offshore vs. local)
• Built scalable pipelines to segment petabytes of customer service data for data anaysis using SQL and R
• Investigated the impact of wait time on customer contact experience to re-evaluate existing service level targets, helping leadership re-evaluate service level targets

As an anlayst for the marketing team of a technology giant, I provided analytics solutions and aided data driven decision making. Highlights include:

• Performed customer segmentation using hierarchical clustering; segment characteristics were used to address smartphone returns
• Identified key demographic and psychographic drivers that lead to Smartphone activations using multiple linear regression models; this helped the client design more targeted marketing campaigns
• Procured daily level device pricing data by web-crawling using Python; aided the client to perform pricing analysis

Mentorship Session

Bijo Thomas🇺🇸

About Bijo Thomas

Expertise

Sessions with Bijo Thomas