Technical Expertise
Generative AI & LLMs
Agentic AI, Advanced RAG, LangChain, OpenAI, Hugging Face, Llama Index, Multimodal Models, Prompt Engineering, Fine-Tuning, Semantic Chunking, Hybrid Retrieval.
Data & ML Engineering
Python, R, SQL, TensorFlow, PyTorch, Keras, Scikit-learn, Pandas, Matplotlib, Data Structures and Algorithms, Text/Image Analysis.
Cloud & MLOps
AWS (EC2, Lambda, S3), Azure Data Warehousing, HPC, CI/CD, MLOps, Prefect, Spark/PySpark, Git/GitHub.
Deep Learning
Transformers, CNNs, RNNs, LSTMs, Deep Reinforcement Learning.
Education
M.S. in Information Science (ML)
University of Arizona
July 2024 - May 2026 | GPA: 4.0
- Machine Learning & Deep Learning
- Applied Natural Language Processing
- Data Mining & Analysis
- Cloud Data Warehousing (Azure)
Bachelor's in Data Science
Jain University
Aug 2021 - July 2024 | CGPA: 9.3/10
- Statistical Inference & Modeling
- Machine Learning Algorithms
- Time-Series Analysis
- Advanced Data Visualization
About Me
Avi Kumar Talaviya is a data scientist and ML engineer passionate about turning complex data into actionable intelligence. With hands-on experience in statistical inference, data visualisation, data analytics tools like pandas, numpy, statsmodels, sklearn, machine learning, deep learning, and large language models, he has applied advanced analytics to domains ranging from healthcare EEG/fMRI research to environmental AQI forecasting and traffic-severity prediction.
Avi is skilled in Python, R, SQL, cloud deployment on AWS, and building end-to-end AI solutions using frameworks like TensorFlow, PyTorch, and LangChain. He enjoys designing scalable pipelines, optimising models on high-performance computing environments, and mentoring learners in data analytics. Avi Kumar Talaviya is suitable for the data science, machine learning and natural language processing data scientist roles and open for such roles in the technology, product and healthcare domains.
Interview Prep ✨ (Powered by Gemini)
Click to generate a challenging interview question based on the experience and projects listed on this page.
Work Experience
Research Collaborator - University of Arizona (ECE)
Aug 2025 - Present | Tucson, AZ
- Designed and deployed data-preprocessing pipelines for EEG-fMRI signals, boosting downstream model stability by 15%.
- Implemented YOLO-based object detection to automate region-of-interest identification in neuroimaging data, cutting manual labelling time by 30%.
- Trained and fine-tuned transformer models on the university’s high-performance computing cluster, reducing runtime by 25%.
- Integrated multi-modal EEG and fMRI features using transformer architectures for an 18% gain in classification accuracy.
Data Specialist - TOPS Technologies
Sep 2024 - Apr 2025 | Surat
- Processed 500GB+ of raw, multi-source data using Python (Pandas, NumPy), reducing data preparation time by 40%.
- Implemented robust validation checks, achieving 99.9% accuracy in mission-critical financial reports.
- Optimized SQL workflows through stored procedures, decreasing query execution time by 60%.
- Analyzed customer data to identify churn drivers, contributing to an 8% reduction in churn.
- Built predictive models using Scikit-learn to forecast demand with 85% accuracy.
Junior Machine Learning Engineer - Omdena
Sep 2022 - July 2023 | Remote
- Developed Decision Tree model to predict high-risk patients, achieving a final Recall of 83%.
- Performed SMOTE upsampling to reduce dataset imbalance, enhancing identification of minority class patients.
- Utilized K-means clustering to segment patients, distinguishing between demographics for targeted analysis.
- Operationalized model predictions by implementing Python-based training and ranking scripts.
Data Science Project Lead - Omdena Mumbai
Mar 2023 - May 2023 | Remote
- Led a team of 25+ members in the development of a predictive model for Air Quality Index (AQI) forecasting.
- Managed analysis and complex preprocessing of time-series data, achieving 90% efficiency increase in model performance.
- Implemented cloud-based data ingestion pipelines using AWS Lambda and S3.
- Developed a time-series forecasting model to predict AQI in Mumbai with over 80% accuracy.
Career Timeline
Research Collaborator
Multimodal EEG-fMRI signal processing & transformer fine-tuning at UofA.
Data Specialist
Enterprise data engineering and BI optimization at TOPS Technologies.
Bachelors in Data Science
Graduated from Jain University with 9.3/10 CGPA.
Omdena Project Lead
Managed cloud infrastructure and time-series forecasting for AQI projects.
Featured Projects
Medical Healthcare LLM Fine-Tuning
NLP-Project Repository
- Fine-tuned an LLM on medical healthcare chat data for domain-specific dialogue.
- Applied PEFT (LoRA) techniques to optimize model weights with limited compute.
- Implemented data cleaning pipelines for unstructured medical chat datasets.
- Evaluated accuracy using ROUGE and BLEU metrics for healthcare intent.
Collaborative Book Recommender
- Built a matrix factorization engine for user-item recommendations.
- Applied dimensionality reduction to handle large-scale rating data.
- Achieved high precision in predicting similar reading preferences.
Cloud Data Warehousing
- Designed a scalable Star Schema for enterprise-level analytics.
- Developed automated ETL pipelines for multi-source data ingestion.
- Integrated BI dashboards for real-time operational reporting.
Cyberbullying Detection NLP
- Processed unstructured text data using word2vec embeddings.
- Trained XGBoost classifier with optimized log-loss performance.
- Resolved class imbalance in toxicity datasets for robust detection.
Traffic Severity Prediction
- Applied Principal Component Analysis (PCA) for feature dimensionality reduction.
- Treated multi-source traffic data for missing values and outliers.
- Categorical encoding for spatial and temporal features.
Research Work
Classifying Sleep and Rest States using EEG-fMRI Fusion Transformer
Multimodal Deep Learning for Healthcare
- Developed a novel multimodal transformer fusion architecture for classifying brain states (sleep/rest) by integrating EEG and fMRI signals.
- Utilized fMRI's submillimeter-to-millimeter spatial resolution for precise localization of neural processes.
- Incorporated EEG's millisecond-level temporal resolution to capture rapid neural dynamics for high-fidelity state tracking.
- Design a strategy for multimodal training (fMRI + EEG) that enables low-cost, portable, and accessible inference using only EEG, maximizing utility and accessibility.