Applied Machine Learning
Build practical prediction systems for classification, forecasting, scoring, and decision support. Includes computer vision with deep learning for image and video analysis.
What we do
We build ML systems that solve real problems - not academic exercises. From tabular prediction models to computer vision with deep learning, we design, train, evaluate, and deploy models that deliver measurable business value. Every model is packaged for production with monitoring and maintenance plans.
Classification
Binary and multi-class classification for fraud detection, risk scoring, customer segmentation, and decision support.
Regression & Forecasting
Predict continuous values and future trends: demand forecasting, price prediction, resource planning.
Feature Engineering
Design and extract meaningful features from raw data. Often the difference between a good model and a great one.
Computer Vision
Image classification, object detection, segmentation, and visual inspection systems using deep learning.
Deep Learning
Neural networks for complex patterns: CNNs for images, transformers for sequences, custom architectures for specialized tasks.
Anomaly Detection
Detect unusual patterns in operations, finance, infrastructure, and business data. Includes risk scoring and alert systems.
What you need to get started
Labeled or historical data
Data with known outcomes for supervised learning, or historical data for unsupervised approaches. More data usually means better models.
Clear prediction target
What should the model predict? A category, a number, a probability, or an anomaly score? Define the output clearly.
Performance requirements
Accuracy, precision, recall, latency, or throughput targets. What trade-offs are acceptable?
Deployment environment
Where will the model run? Cloud, on-premise, edge device, or mobile? This affects model size and architecture choices.
What we've built
Golden Gauss AI - ML System
Gradient boosting model with 239 engineered features for gold market prediction. Includes feature engineering pipeline, model training and evaluation, ONNX packaging for deployment, and probability-based decision system.
Vacant Lot Detection
Computer vision model for object detection and image segmentation on satellite imagery. Built for the Solafune competition, detecting vacant lots in urban areas using deep learning segmentation architectures. Ranked Top 8% (32 / 447).
FlexTrack Challenge - Energy Flexibility
Hybrid two-stage ensemble framework for detecting and quantifying energy flexibility in buildings via demand response events. Stage 1 classifies DR events using LightGBM + XGBoost ensembles; Stage 2 regresses energy impact. G-Mean: 0.618, nMAE: 0.991.
GeoFM Crop Classification
Competition solution for NASA/Amini crop classification using geospatial foundation models, satellite time-series, PyTorch, and LightGBM ensemble.
Golden Gauss AI - ML System
Gradient boosting model with 239 engineered features for gold market prediction. Includes feature engineering pipeline, model training and evaluation, ONNX packaging for deployment.
Short Activist Predictor
Financial text classification using BERTopic and scikit-learn. Published Python package with Hugging Face model and CI/CD pipeline.
ATLAS Benchmark
AI evaluation benchmark measuring learning ability across multiple faculties. Includes replay viewer, adaptation testing, and failure mode analysis.
Have a prediction problem to solve?
Tell us about your data and what you need to predict.