Nikhil Singh

Spearheaded the development of EVA ChatBot, a Retrieval-Augmented Generation (RAG) model, to summarize over 200+ complex reports by integrating vector search with a Hybrid approach, combining Naive RAG and knowledge Graph RAG techniques.

Engineered a Self-Agentic RAG architecture featuring intelligent query refinement, metadata-aware retrieval, and automated fact verification, significantly enhancing contextual relevance and accuracy.

Designed and implemented a Query Refinement and Fact-Checking Langgraph agent, improving factual accuracy by iteratively refining weak queries and validating retrieved content against metadata (geography, sector, report type).

Conducted rigorous model performance evaluation using REHL, RAGAs, and DEEPeva metrics to ensure accuracy across multiple dimensions.

Led the development of an AI-driven scoring system for the APTIS for Teens writing task, automating language proficiency evaluation using Machine Learning and Large Language Models (LLMs).

Utilized the Langchain framework to generate robust features for assessing writing and speaking skills, and built custom word list features, resulting in a 25% improvement in scoring precision for language models.

Developed and implemented an Aspect-Based Sentiment Analysis (ABSA) project to evaluate social media opinions regarding the British Council and UK universities.

Applied advanced models including VADER, BERT, RoBERTa, and Small Language Models (SLM) to analyze sentiment towards entities and identify underlying reasons.

Optimized token limits through key phrase extraction and topic modeling, employing a cost-efficient approach to uncover primary factors driving strong sentiments, validated through human scoring.

Developed and implemented advanced Statistical and Machine Learning models to forecast demand, consistently outperforming traditional demand planners and enhancing supply chain decision-making.

Achieved a 10% average improvement in forecast accuracy by developing innovative features that effectively addressed high intermittency in diverse product categories.

Integrated critical factors such as seasonality, promotions, holidays, price changes, and weather patterns into demand forecasting models using sophisticated statistical logic.

Led multivariate forecasting initiatives leveraging a diverse range of techniques, including tree-based models, linear models, and deep learning architectures.

Developed and deployed Statistical and Machine Learning models for time series forecasting, utilizing univariate models (e.g., ARIMA) and Tree-based models for Intermittent Demand of failure parts in aircraft.

Created robust ML models to accurately estimate Remaining Useful Life (RUL) of critical components, significantly enhancing predictive maintenance capabilities.

Performed comprehensive Reliability Modelling for parts failure time and optimized parameters for underperforming models, leading to improved system reliability.

Designed and implemented key metrics for root cause analysis, facilitating effective corrective actions on defective parts and improving overall operational performance.