Rahul Kumar

Led AI roadmap for resolving fleet management queries using ReAct-based AI agent, reducing turnaround time by 28%.
Architected and designed Fuel Optimization Platform using K-Means, XGBoost, Uplift Model, and SHAP explainability, cutting fleet fuel consumption by 10%.
Built LLM-powered SQL agent with domain-specific knowledge graphs, improving accuracy by 30% and reducing errors by 18%.

Led AI roadmap for resolving fleet management queries using ReAct-based AI agent → reduced turnaround time by 28%., Architected and designed Fuel Optimization Platform (K-Means + XGBoost + Uplift Model, SHAP explainability) → cut fleet fuel consumption by 10%., Built LLM-powered SQL agent with domain-specific knowledge graphs → improved accuracy by 30%, reduced errors by 18%.

Drove and designed pretrained and finetuned NLP language models (T5) for complaint identification and sequential models (LSTM) for complaint or contact propensity, reducing support costs by 15%.
Built product and feature detection two-stage architecture pipeline with fine-tuned transformers (Longformer, SetFit), boosting accuracy from 72% to 89%.

Drove and designed Pretrained & Finetuned NLP language models (T5) as well as Sequential Models (LSTM) for complaint identification and Complaint or Contact Propensity respectively → reduced support costs by 15%., Built product and feature detection two-stage architecture pipeline with fine-tuned transformers (Longformer, SetFit) → boosted accuracy from 72% → 89%.

Forecasted cell tower traffic using neural networks, reducing energy waste by 20%.
Architected an AI-driven monitoring system using time-series models for real-time anomaly detection and to predict order failures with high accuracy.
Automated root cause analysis using Dynamic Time Wrapping (DTW).
Operationalized ML-powered insights via a Flask API.
Cut down RCA turnaround time by 50%.
Delivered content recommender system, improving personalization and coverage by 25% through A/B testing.

Forecasted cell tower traffic using neural networks → reduced energy waste by 20%., Architected an AI-driven monitoring system using time-series models for real-time anomaly detection and to predict order failures with high accuracy. The system automated root cause analysis using Dynamic Time Wrapping (DTW) and was operationalized by serving all ML-powered insights via a Flask API. It cuts down RCA turnaround time by 50%., Delivered content recommender system → improved personalization & coverage by 25% (A/B tested).

Engineered an advanced feature engineering framework using Bayesian inference to create highly reliable, adjusted ratios from sparse data.
Used probabilistic ratios as robust inputs for downstream machine learning models, such as supplier quality ranking systems.
Validated the framework through rigorous backtesting.
Achieved a significant uplift in Mean Average Precision (MAP) over traditional metrics.

Engineered an advanced feature engineering framework using Bayesian inference to create highly reliable, adjusted ratios from sparse data. These probabilistic ratios served as robust inputs for downstream machine learning models, such as supplier quality ranking systems. The framework's superiority was validated through rigorous backtesting, which showed a significant uplift in Mean Average Precision (MAP) over traditional metrics.