3. Data, Science & AI

Parses power system network data including buses, generators, and branches for power flow analysis.

Executes DC power flow analysis for transmission networks, including susceptance matrix construction, line flow calculation, and sensitivity analysis.

Designs and implements scalable data pipelines, ETL/ELT systems, and real-time streaming infrastructure with modern data stack technologies.

Extracts frames from video files and saves them as image files using the OpenCV computer vision library.

Analyzes images and multi-frame sequences using OpenAI's multimodal GPT models for visual content understanding.

Optimizes power generation costs by computing optimal generator setpoints, calculating operating margins, and handling generator cost functions in power systems.

Extracts text content from images using Tesseract OCR engine via Python, enabling automated data capture from visual sources.

Provides foundational guidance for running the General Lake Model (GLM) to simulate lake temperature dynamics, including input file structure and parameter configuration.

Analyzes videos using Google Gemini API for object counting, pedestrian detection, and vehicle tracking in surveillance applications.

Performs Difference-in-Differences causal analysis to identify demographic drivers of behavioral changes with p-value significance testing for A/B tests and policy evaluation.

Automates cleaning of messy tabular data including deduplication, missing value imputation, outlier handling, and text normalization.

Enables fuzzy search for fund and stock information, allowing accurate retrieval despite misspelled or incomplete names in financial datasets.

Implements Manifold-Constrained Hyper-Connections (mHC) algorithm for stabilizing deep neural network training using Sinkhorn-Knopp on doubly stochastic matrices.

Enables spatial analysis of hexagonal grid maps for Civilization 6 strategy optimization and pattern recognition.

Efficiently trains GPT-2 scale language models (124M parameters) on a single GPU with mixed precision and modern optimizers like AdamW.

Transforms raw data into ML-ready features through categorical encoding, numeric scaling, and feature selection.

Analyzes images and multi-frame sequences using OpenAI GPT vision models for computer vision tasks.

Provides a manifold-constrained hyper-connections algorithm to stabilize deep neural network training using doubly stochastic matrices.

Automates dataset preparation for ML by encoding, scaling, and selecting features to improve model performance.

Strategy for optimizing item placement on spatial maps to maximize objectives under constraints.

Fuses visual (face/lip) and audio features via OpenCV and embeddings to enhance speaker diarization accuracy in video files.

Transcribes audio to text using Whisper models (small, base, medium, large-v3) or faster-whisper, aligning timestamps with diarization for speaker-labeled subtitles.

Enables running Python-based ML model training on cloud GPUs (A100/T4/A10G) via Modal, handling setup, GPU selection, and data management.

Fuses visual cues (face/lip tracking via OpenCV) with audio embeddings to enhance speaker diarization accuracy in video files with multiple speakers.