Results-driven, analytically strong **Data Scientist and Machine Learning Engineer** with hands-on experience building, optimizing, and applying machine learning models for complex, data-constrained problem domains. Skilled in statistical analysis, **A/B testing**, **predictive modeling,** and **AI-driven automation, with a strong ability to convert complex data into actionable insights**. Worked as an **intern at Channel Science** on a high-impact U.S. nuclear monitoring–aligned project focused on **classifying earthquakes versus nuclear explosions** using seismic wave patterns (P-waves vs. S-waves). Contributed to **decoding and digitizing legacy seismographic data** stored on magnetic tapes, necessitated by the Nuclear-Test-Ban Treaty restrictions on new data collection. Applied **sequence-to-sequence models for seismic signal decoding** and **autoencoder-based anomaly detection** to distinguish natural seismic events from explosion signatures. The work was supported by an SBIR Grant (DE-SC0021879) and was recognized with a **LinkedIn recommendation from the Technology Leader**. Won Hackathons, published papers, and Won for presenting a paper in a conference involving **Machine Learning, Generative AI,** and **Agentic AI** to drive innovation and strategic growth.
Results-driven, analytically strong **Data Scientist and Machine Learning Engineer** with hands-on experience building, optimizing, and applying machine learning models for complex, data-constrained problem domains. Skilled in statistical analysis, **A/B testing**, **predictive modeling,** and **AI-driven automation, with a strong ability to convert complex data into actionable insights**. Worked as an **intern at Channel Science** on a high-impact U.S. nuclear monitoring–aligned project focused on **classifying earthquakes versus nuclear explosions** using seismic wave patterns (P-waves vs. S-waves). Contributed to **decoding and digitizing legacy seismographic data** stored on magnetic tapes, necessitated by the Nuclear-Test-Ban Treaty restrictions on new data collection. Applied **sequence-to-sequence models for seismic signal decoding** and **autoencoder-based anomaly detection** to distinguish natural seismic events from explosion signatures. The work was supported by an SBIR Grant (DE-SC0021879) and was recognized with a **LinkedIn recommendation from the Technology Leader**. Won Hackathons, published papers, and Won for presenting a paper in a conference involving **Machine Learning, Generative AI,** and **Agentic AI** to drive innovation and strategic growth.
- Worked on project funded by the US Department of Energy, SBIR Grant Award: DE-SC0021879
- Implementing a **sequence-to-sequence classification model** for seismographic signal decoding, achieving **98% accuracy** and an **MCC of 96%** for sequence tagging.
- Processed recorded magnetic tapes by **segregating tracks into clusters**, aligning signals, and correcting skew using **Dynamic Time Warping (DTW)** techniques.
- Developed **anomaly detection models** with autoencoders, achieving **94% accuracy**, and performed **data visualization and statistical analysis** on channel signals based on sync points and record lengths.
- Developed an **Agentic AI–based skincare recommendation system** using **LangGraph** to orchestrate end-to-end workflow execution.
- Implemented **skin type classification** aligned with **Bauman’s Skin Type system** using the **CLIP Visual Language Model (VLM), Parallel RAG architecture using LangGraph** for image–text understanding, and implemented a CNN for age and gender with an Accuracy of 85% and Mean Absolute error (MAE) of 0.36 using the UTKFace dataset.
- Optimized response time by **parallelizing multiple RAG pipelines**, significantly reducing overall inference latency.
- Designed and deployed **two recommendation pipelines**:
- a **GenAI-driven product recommendation engine**,
- and a **web-scraping–based system** that retrieves product information directly from official brand websites based on the identified skin type, age, and gender.
- Conducted **A/B testing** to evaluate recommendation quality, comparing **ChatGPT-based outputs** with the custom-built recommendation system with a **Z-value of 6.19** and **p-value of 5.9*e^{-10}$** as well as benchmarking performance across the two internal pipeline versions.
- Conducted **extended log analysis for MF4 files** with Tata Motors Limited, focusing on **SRR features** such as **Rear Cross Traffic Alert (RCTA)**, **identifying** **anomalies,** and providing detailed reports.
- Applied **Python-based machine learning and data visualization**, achieving an **F1 score of 92%,** significantly improving the accuracy and reliability of the outputs of the **Door Open Alert (DOA) feature**.
- Developed a **sensor-based data collection application** in Python for real-time handling and analysis of sensor data, optimizing it for machine learning pipelines.
- Addressed **false warnings in RCTA 45-degree parking** through **Feature engineering** by mathematical modeling, improving detection accuracy from **78% to 85%** for Fiat Chrysler Automobiles (FCA), earning **recognition from the Branch In-Charge**.
- Designed and administered Oracle database solutions to enhance the retrieval and preprocessing of data.
- Analyzed over four years of tender data and Improved tender win rate by **18%** while maintaining an average **12% higher profit margin.**
- Analyzed sentiment analysis model efficiency with labeled datasets to optimize classification for customer feedback into actionable analytics.
- Conducted **A/B testing** for the **New India Assurance Mediclaim Policy** to find its efficiency and validity strategies of **confidence Interval of 95%**.