Arjun Darbha

Data Scientist at Seraph Biosciences, Inc

Optimized a PyTorch model to predict bacterial strains from Raman spectra, achieving F1 of 0.96 through Optuna hyperparameter tuning. Ported MATLAB data pipelines to Python, applying SciPy signal processing to reduce noise in raw spectral data.

Principal Data Scientist at Deepthought AI (Illumety Inc & DVS Baseball)

Led end-to-end AI and data-platform initiatives, architecting a video-to-metrics workflow using Azure Durable Functions, FastAPI microservices, and Azure Cosmos DB to onboard 3,000 clients across 30 facilities. Built real-time data streaming pipeline with Apache Pulsar and Azure Functions for news prioritization; developed a Python library for CRUD on AstraDB (Cassandra) and Elasticsearch to power web-app services. Implemented a high-performance tree-based ensemble model for Raman spectra classification using Azure AutoML, achieving Recall 1.0 and Precision 0.92. Created a time-series application with MACD to surface high-risk messages, deployed via Azure Functions. Led product vision and engineering roadmap, delivering AI-driven analytics platform and MLOps pipeline; managed cross-functional teams, reducing model deployment time by 40%.

Principal Data Scientist at Deepthought AI

Architected real-time data pipelines using Apache Pulsar and Azure Functions to prioritize news and surface high-risk messages, improving data freshness by 30%. Engineered end-to-end MLOps workflows with Azure Durable Functions, FastAPI, and Azure Cosmos DB, enabling scalable video-to-metrics processing that secured 30 new facilities and onboarded 3,000 clients. Developed high-performance tree-based ensembles and anomaly detection models (One-Class SVM, Isolation Forest) on Raman spectra, achieving recall 90% on highly imbalanced datasets. Built a custom Python library for CRUD operations on AstraDB (Cassandra) and ElasticSearch, streamlining data access for web-app services. Led product vision and engineering roadmap, coordinating cross-functional teams to deliver AI-driven analytics solutions for consumer platforms. Designed and deployed machine learning pipelines for real-time joint-coordinate detection in video streams, reducing inference latency by 25%. Oversaw cloud infrastructur

Data Science Consultant at EatSleepRIDE

Built a rider safety score model using telematics data, adopted by insurers for underwriting decisions. Developed a deep learning model to detect motorcycle maneuvers with 0.92 F1-score and engineered an end-to-end TensorFlow pipeline deployed as TensorFlow Lite and CoreML for mobile apps.

Data Scientist at EatSleepRIDE

Implemented a rider safety score model using te­lematics data, now used by insurers for underwriting decisions. Developed a deep-learning maneuver detection model with 0.92 F1-score, integrated into mobile apps via TensorFlow Lite and CoreML.

Data Scientist at Aptiv

Designed root-cause analysis tools using PySpark MLlib and XGBoost to identify sensor failure probabilities, saving millions in warranty costs. Developed a trip-score metric quantifying driving behavior, providing actionable insights for fleet management.