Hi! I'm Arun
Data Scientist | AI/ML Engineer | Data Analyst | Data Engineer
Passionate about solving problems using data and AI.
5 years of experience of turning real world data into actionable insights and solutions.
Summary
With 5 years of experience spanning technology, mining, and manufacturing industry, I am a data science and engineering professional. With Master’s degree in Data Science from Rutgers University and dual degree in Engineering from IIT Dhanbad, I am proficient in Python, R, SQL, and modern data tools and bring hands-on expertise across Data Science, Artificial Intelligence & Machine Learning, Data Analysis, Data Engineering, that drive revenue growth and optimize critical business metrics. I thrive in cross-functional environments, transforming complex data into actionable insights and leading projects from ideation to deployment. I possess strong leadership & team coordination skills, analytical thinking, clear communication, adaptability, and a collaborative approach. My business acumen enables me to align data-driven solutions with strategic objectives and deliver measurable value for the organisation. Curiosity-driven and committed to continuous learning, I’m passionate about using data and technology to solve real-world problems and contribute to high-impact, innovative initiatives.
LANGUAGES
Python
SQL
R
JavaScript
TypeScript
C++
HTMLMachine Learning
Pandas
Numpy
Matplotlib
Tensorflow
PyTorch
Sci-kit Learn
Seaborn
Hugging Face
KerasGenerative AI
OpenAI
LLama2
Mistral
Langchain
Natural Language Processing
Retrieval Augmented Generation
Agentic AI
Vector DBDATABASES
PostgreSQL
MySQL
MongoDB
DuckDB
Elastic Search
SnowflakeINFRASTRUCTURE
AWS Cloud Services
Azure Cloud Services
Google Cloud Platform
Docker
Kubernetes
GitHub
GitLabTOOLS
Visual Studio Code
Power BI
MS office
ChatGPT
GitFRONTEND
CSS
React
Three jsBACKEND
Django
Node.js
Next.jsWork Experience
- Designed an accessibility-first AI chatbot for Deaf users integrating Natural Language Processing (NLP) and real-time text-to-ASL translation, ensuring WCAG-compliant UI and culturally relevant content.
- Led cross-functional discovery sprints applying agile frameworks (Scrum, Kanban) to coordinate AI engineers, UX researchers, and Deaf educators, facilitating data engineering for sign-language datasets for the MVP.
- Automated reagent dosing in froth flotation circuits using Random Forest regression models, resulting in a 2.1% increase in lead recovery and 2.5% in zinc, which contributed to $5.9M in additional annual revenue.
- Engineered and processed over 1.5 million rows from 106 data sources employing SQL and R to reduce power consumption by 2%, leading to $70,000 in yearly savings.
- Designed and maintained scalable ETL pipelines and Power BI dashboards to monitor daily production metrics, equipment downtime, and maintenance spend, improving operational visibility and real-time decision-making in daily leadership meetings.
- Defined and tracked key utilization metrics for Heavy Earth Moving Machines (HEMM) and Ball Mills, identifying inefficiencies that reduced contractor costs by $144,000 annually through data-driven asset reallocation.
- Applied Kanban agile practices in collaboration with global teams across various time zones, ensuring timely and effective project delivery.
- Implemented process improvements in the water workflow that reduced freshwater consumption by 9.5% (from 85m³ to 77m³ per hour), maintaining plant throughput while ensuring regulatory compliance with environmental guidelines.
- Led a team of 40+ skilled and unskilled workers as Shift Engineer and In-Charge during the successful commissioning of a 2.0 Million Tonnes per Annum processing plant
- Collaborated with international vendors to oversee on-site lab trials on company ore/processes; collected and analysed data, and presented findings to the senior leadership and General Manager. Upon approval, led full-scale plant trials to implement and evaluate vendors' solutions.
- Analyzed water consumption data across 2,040 eight-hour shifts spanning two years, and developed an interactive Power BI dashboard that enabled operations teams to identify inefficiencies and optimize water usage in the coal beneficiation plant.
- Performed detailed water balance calculations across three critical plant sections using industry-standard hydrological models, leading to the identification of key areas for water conservation to ensure government environmental regulations compliance.
- Developed robust ETL pipelines and Power BI dashboards to visualize daily, monthly, and yearly production data, enabling operations and leadership teams to monitor performance trends and make data-driven decisions.
- Maintained and optimized existing data pipelines, troubleshooting ingestion and data quality issues to ensure high availability, accuracy, and efficiency across the plant’s reporting workflows.
Selected Projects
Domain-Specific Medical LLM Fine-Tuning using LoRA & Scalable QA Pipeline
Engineered a scalable, parameter-efficient fine-tuning pipeline for Llama 2 using LoRA adapters, achieving faster training and lower memory usage on GPUs. Implemented custom instruction-style prompting and leveraged float16 precision to further accelerate training, reduce inference time significantly (half) and memory usage on GPUs, while maintaining high accuracy with minimal precision degradation.
Fraud Detection with a novel math-driven oversampling strategy outperforming ADASYN, SMOTE and Tabnet (Transformer Architecture from google for Tabular Data)
Developed a novel math-driven oversampling strategy that intelligently identifies and amplifies hard-to-classify samples using a custom probability-based scoring system. Engineered dynamic, adaptive sampling that targets rare and difficult instances boosting the model's focus where it matters most. Tuned class balance with precision through adaptive lambda scaling, avoiding dataset bloat. Rigorously benchmarked against SMOTE, ADASYN, TabNet, and the base model, consistently outperforming them across metrics with far fewer oversamples. Ensured robust evaluation with strict test set isolation, avoiding data leakage. This solution delivers sharper classification by strategically reinforcing model weaknesses.
AI-Driven Optimization for Maximizing Metal Recovery in Mining
Optimised reagent dosing in iron ore flotation by using Random Forest model, automating the process while minimising chemical waste. Engineered and refined a large-scale industrial dataset (580,000+ records, 29 features), conducting rigorous feature selection using OLS p-values and multicollinearity diagnostics. Delivered a robust, data-driven solution that enhanced process efficiency and supported sustainable extraction.
Sentiment Analysis of Popular Songs' Lyrics
Analysed over 169,000 top Spotify tracks to uncover evolving trends in music sentiment and audio attributes using R. Conducted in-depth exploration of features highlighting how musical composition shifted across decades and artist eras. Delivered key findings showing modern songs trend toward higher energy, louder volumes, and increased danceability, while exhibiting lower acousticness and instrumentation.
Time Series Analysis with Conformal Prediction
Benchmarked traditional ARIMA against the state-of-the-art TSDiff diffusion model for probabilistic stock price forecasting using Yahoo Finance data across various tickers. Engineered a forecasting pipeline to evaluate model accuracy and trend-tracking capabilities on highly volatile time series. Leveraged ARIMA with dynamic retraining and TSDiff with quantile-guided DDPM sampling, generating 100 probabilistic samples per prediction. Analysed performance via MAE and MSE, revealing ARIMA's superior precision on short-horizon forecasts. Delivered key insights into the limitations of diffusion-based models on non-periodic financial data, while showcasing practical strengths of classical statistical methods.
Leadership, Honors and Extracurricular Activities
*
Achieved 98.7 percentile in the Common Admission Test (CAT) 2022 among 300,000 candidates, earning interviews from India’s top management institutes.
@ Common Admission Test-2022
*
Ranked in the top 0.8% out of 1 million candidates in the IIT-Joint Entrance Examination 2013, earning admission to IIT Dhanbad.
@ IIT-Joint Entrance Examination(Advanced)-2013
*
Elected as Placement Representative and Coordinator, Student Body, IIT Dhanbad. Elected person represents his entire class in Training and Placement (Job) cell of the college.
@ IIT Dhanbad
*
Founded "Vakta [Orator]", a Toastmasters-inspired club at IIT Dhanbad, empowering 100+ students in public speaking and interview readiness.
@ IIT Dhanbad
*
Completed 500+ hours of active practice in presentations, public speaking, and body language over four years, resulting in significant improvement in communication skills.
@ IIT Dhanbad
*
Community Service: Volunteered as a Math Teacher with "Kartavya [Duty]" (2014-2017), supporting over 50 students at one of largest student-run NGO in India.
@ IIT Dhanbad
