Arpit Rathi

Machine Learning Specialist at Startups [Entrepreneurship + Freelance] (2023-04 – Present)

Founded an AI-powered note-taking app for project knowledge management, led team assembly, formulated business model, conducted user, product, & market research, designed UI/UX wireframes, and crafted a pitch deck.
Led the development of PoCs for NLP pipelines, including semantic search, zero-shot classification, paragraph ordering, and auto-tagging.
Architected a PoC for an end-to-end streaming data pipeline with Kafka, Zookeeper, Schema Registry, Avro, and Spark Structured Streaming for real-time processing of network data, reducing latency by 90%, and developed an HTTP tunneling attack detection pipeline with 0.95 F1-score.
Prototyped a content recommendation system optimizing for relevance and diversity, for a B2C application focused on mental health & wellness.

Lead Data Engineer (Machine Learning) at Oliver Wyman [B.U. – Data Monetization Unit] (2022-06 – 2023-03)

Conceived and pitched three innovative product ideas—usage-based insurance, credit portfolio monitoring, and investment risk appetite—forecasting a combined revenue potential exceeding $50M; conducted product development research, crafted persuasive pitch decks, and secured stakeholder buy-in.
Developed a PoC for driver-risk scoring, integrating driver telematics and credit data using PySpark, and achieved 5–7x lift in claim frequency predictions at individual and aggregated levels, enabling accurate risk stratification.
Performed rate vs. mix attribution analysis using Experian's credit database, collaborated on module design & development for advanced climate credit risk modeling, and cloud deployment of the model validation toolkit.

Manager - Machine Learning at Heal Software [Appnomic Systems] (2020-04 – 2022-04)

Spearheaded the ML R&D for AIOps solutions, including anomaly detection, capacity forecasting, alert correlation, root cause analysis, and preventive action recommendations.
Reduced alert volumes by up to 70% by implementing a logistic auto-regressive model for causal dependency graph mining from multivariate time-series data and grouping correlated alerts during real-time inference.
Implemented Regression-Enhanced RF and XGBoost models for efficient extrapolation in IT infrastructure resource capacity forecasting, achieving a 60–80% reduction in RMSE scores for high-throughput scenarios.
Improved performance by 50% in a clustering-based implementation of the patented load-behavior correlation methodology for outlier detection by introducing a probabilistic regression model.
Designed a novel online outlier detection algorithm that reduced memory utilization by over 90% and eliminated frequent model retraining requirements, thereby enhancing system efficiency and scalability.
Refactored the core ML module (25+ classes) to align with Scikit-Learn standards, introduced state-of-the-art ML algorithms, and a model-agnostic data preprocessing pipeline for workflow optimization.
Advanced the company's intellectual property portfolio by filing three patent applications — an end-to-end proactive healing system, an application performance and customer experience scoring system, and an accelerated anomaly detection system for new cloud environments.

Machine Learning Engineer at Emtropy Labs [Support Logic] + Freelance (2018-01 – 2020-03)

Fine-tuned BERT transformer-based models for multi-label classification, achieving an AUC score above 0.90; implemented aspect-based sentiment analysis of professional competencies using peer-to-peer feedback data, along with an XAI model, and deployed the models on the cloud.
Established the product foundation by designing a novel methodology and implementing a PoC for a customer service QA automation solution, extracting customer intent & emotion, and corresponding behavioral competencies from CSR responses for objective assessment of agent performance while ensuring solution explainability.
Delivered impactful solutions for synthetic data generation for automated number plate recognition, Bayesian networks-based anomaly detection on VLSI systems design data, and toxic content classification projects.

R&D Engineer (Machine Learning) at Next Education India Pvt. Ltd. (2017-07 – 2017-11)

Engineered PoC for an automated grammatical error correction tool for K-12 students, achieved 80% accuracy on preposition & determiner prediction tasks using contextual feature extraction and classification.

About