Divya Prakash

Created a scalable, agent-based document quality review system using LangGraph, orchestrating over 35 compliance checks via a graph-driven architecture. Designed an algorithm to map LLM outputs to exact spans in Word and PDF files, enabling precise, non-invasive feedback and automated annotations. Achieved 90% recall and 70% precision, cutting review time for lengthy contracts from days to hours while enabling human-in-the-loop workflows.

Developed supervised ML models on over 10M financial records to predict critical corporate events like M&A, bankruptcy, and delisting, achieving a top-25 precision of 85%. Implemented k-year fold training and ensemble techniques to address temporal data drift, improving both model accuracy and business KPIs. Optimised training pipelines on Azure Databricks, cutting iteration time from days to hours. Technology used included Python, Scikit-learn, Optuna, Hydra, Databricks, Numpy, RNN, LSTM.

I developed advanced deep learning algorithms focused on bias detection and content ranking. My work involved building robust models capable of identifying polarising news articles, contributing directly to the detection of misinformation and bias in digital contents. I secured top ranks in the highly competitive SemEval NLP challenges (Tasks 4 & 8). This work was driven by a strong technical foundation in Python, PyTorch, Scikit-Learn, Spacy, for NLP and combining traditional and deep learning algorithms.

I Created algorithms to model vehicle fuel profiles and generate actionable insights on fuel usage, wastage, and savings. Developed driving behaviour detection models using OBD-II sensor data and multivariate signal analysis. Granted a patent for a safety driving scoring system, highlighting innovation in vehicle telematics. Tech stack included Python, R, unsupervised modelling, recommendation systems, and advanced sensor signal processing.

I developed real-time analytics for tennis performance by processing smartwatch motion sensor data across six dimensions. I created supervised machine learning models to classify racket swings and deliver actionable player feedback. Applied techniques such as feature engineering, feature selection and regularisation to achieve state of the art performance. Gained proficiency in data visualisation and R along with experience in building end to end system from data collection to prediction and feedback.