L
Lok Venkatesh Vasamsetti
About
Detail
Massachusetts, United States
Résumé
Jobs
verified_user
0% verified
Data ScientistToastAug 2025 - Current (1 year 1 month)• Built TensorFlow demand forecasting models using Python, PostgreSQL, POS orders, reservations, weather, and seasonality signals, improving restaurant forecast accuracy by 18% • Designed Spark ETL workflows feeding AWS SageMaker training pipelines, standardizing high-volume transaction data preparation and reducing model readiness time by 24% • Managed MLflow-based LSTM experiment tracking across restaurant locations, validating accuracy, stability, and cohort-level performance before production deployment • Deployed FastAPI and Docker inference services for real-time restaurant demand prediction, reducing batch forecasting turnaround by 29% during peak service periods • Automated Airflow monitoring and retraining workflows to detect data- EData Science Co-opEntropy LifestyleJan 2025 - Jul 2025 (7 months)• Built a personalized health RAG assistant using LangChain, Claude API, pgvector, FastAPI, and AWS Lambda, grounding responses in user metrics and health scoring logic • Designed LangGraph agent workflows to route health queries through retrieval, scoring, reasoning, and response generation, improving consistency across personalized wellness interactions • Implemented LLM-as-a-Judge evaluation pipelines measuring faithfulness, relevance, and safety alignment, reducing unsupported AI responses through regression alerts and prompt-quality checks • Developed PyTorch sequence models for reorder prediction and health-behavior trends, improving forecasting stability by 18% across food and wellness engagement workflows • Integrated Kafka event st
- Data ScientistGROWWAug 2022 - Jul 2023 (1 year)• Analyzed investor transaction behavior using Python, Pandas, and MySQL, identifying portfolio engagement patterns that improved recommendation relevance by 17% across retail investment journeys • Engineered predictive features from customer demographics, investment frequency, and portfolio history, strengthening XGBoost recommendation models while reducing feature preparation effort by 26% through reusable pipelines • Calibrated classification models with Scikit-learn hyperparameter tuning, balancing recommendation precision and recall before validating investment suggestions through offline evaluation experiments • Conducted A/B testing on recommendation strategies, measuring customer interaction behavior and increasing investment produc
- CData AnalystConTecHubJan 2021 - Jul 2022 (1 year 7 months)• Consolidated operational datasets using SQL and PostgreSQL, producing standardized analytical tables that improved reporting consistency by 22% across customer performance monitoring activities • Investigated customer behavior through exploratory data analysis with Pandas and NumPy, uncovering retention patterns that supported targeted operational improvement initiatives across business teams • Automated recurring Python-based data preparation and validation workflows, decreasing manual reporting effort by 34% while maintaining consistent quality checks before dashboard publication • Created interactive Power BI dashboards summarizing customer acquisition, retention, and operational performance, enabling stakeholders to review business tr
Education
verified_user
0% verified
- Master of Science, Information TechnologyNortheastern UniversityAug 2023 - Dec 2025 (2 years 5 months)
- Bachelor of Engineering, Electronics and CommunicationManipal UniversityAug 2018 - Jul 2022 (4 years)
Projects (professional or personal)
verified_user
0% verified
- PPrecipitation ForecastingPyTorch, LSTM, ConvLSTM2D, Feature Engineering• Developed hybrid PyTorch forecasting models combining LSTM and ConvLSTM2D architectures, improving precipitation prediction accuracy by approximately 11% against baseline sequence models • Engineered multi-source weather signal features using FFT smoothing and convolution techniques, improving heavy rainfall detection consistency across imbalanced environmental datasets • Optimized weighted-loss training workflows reducing no-rain prediction bias, strengthening precipitation event detection across highly skewed meteorological sensor datasets
- PGruntKill – Agentic AI Developer Workflow AutomationPython, FastAPI, Claude API, AWS Lambda, Amazon SQS, pgvector, React• Built an agentic AI workflow assistant using LangGraph and Claude API to break developer requests into planning, retrieval, execution, and validation steps. • Implemented RAG pipelines with pgvector to retrieve project documentation, code context, and workflow instructions, improving grounded responses during developer task automation • Developed FastAPI services with AWS Lambda and Amazon SQS to process asynchronous AI tasks, manage workflow state, and support scalable agent execution • Designed React-based interfaces for submitting developer tasks, reviewing AI-generated outputs, and tracking workflow progress across automated engineering support activities
- LAI Research AssistantLangChain, OpenAI API, MLflow, AWS Lambda, pgvector• Built retrieval-augmented generation pipelines using LangChain and pgvector, improving grounded response relevance during internal document retrieval benchmarking evaluations • Configured semantic chunking workflows replacing fixed-size document segmentation, reducing inaccurate retrieval responses by 14% during iterative testing across enterprise knowledge datasets • Implemented LLM evaluation and prompt optimization workflows using MLflow experiment tracking, strengthening retrieval quality monitoring and model performance validation • Deployed serverless RAG inference services through AWS Lambda, supporting scalable document querying workflows with automated monitoring for retrieval quality degradation detection