Representative Case Studies
 Representative Case Studies

Representative Case Studies

🚘 01. Autonomous Driving – Real-Time Map Fitting

Problem

Curve fitting for autonomous driving map data was slow and storage-heavy, relying on RMS-based polyfit methods.

Solution

We reformulated the problem using parametric boundary representation and derived an analytic Bezier-based formula.

Impact / Result

  • ⚡ 100× faster computation
  • 📉 60× smaller data size
  • 🏆 Led to multiple patents, including one granted in the U.S. and 3 additional patents filed in Korea

Tech Stack

Python, Bezier curve modeling, Custom optimization algorithms

⛴️ 02. Fuel Forecasting for Maritime Shipping

Problem

Captains relied on instinct to balance fuel use and contractual deadlines, resulting in inefficiencies.

Solution

We built a machine learning model based on ship dynamics and weather conditions, with custom features and interpretable outputs.

Impact / Result

  • 🎯 Forecast accuracy within 3–5%
  • 📊 Enabled data-driven routing decisions
  • 🌍 Reduced fuel consumption and emissions

Tech Stack

Python, Scikit-learn, Pandas, Weather APIs, PyTorch, PyTorch, XGBoost, PatchTST

🏠 03. Predicting Real Estate Trends

Problem

Traditional models failed to anticipate turning points or saturation in housing markets.

Solution

We combined global housing datasets with non-linear, non-stationary modeling to capture market dynamics.

Impact / Result

  • 🔮 Predicted upcoming price stagnation and reversal
  • ✅ Matched real-world changes with <10% deviation
  • 💡 Helped decision-makers act before market shifts

Tech Stack

Python, Time-series forecasting, Non-linear regression models,

⚙️ 04. Early Fault Detection for Industrial Systems

Problem

Failures in machinery were only detected post-breakdown, leading to reactive maintenance and delays.

Solution

We developed a real-time anomaly index using sensor features and a custom active learning framework.

Impact / Result

  • 🚨 Pre-failure detection capability
  • ⏱️ Enabled real-time alerts
  • 🔧 Reduced maintenance downtime
  • 🏆 Resulted in a filed patent in Korea for early fault detection

Tech Stack

Python, TensorFlow, Real-time streaming (Kafka), Sensor data processing, Active learning algorithms, Statistical modeling

📢 05. Influencer Recommendation System

Problem

No established metrics or features existed to evaluate influencer effectiveness.

Solution

We created new statistical features and built a recommendation system with a lightweight admin interface and API integration.

Impact / Result

  • 📈 Allowed brand managers to identify high-impact influencers
  • ⚡ Fast prototyping with real-world deployment

Tech Stack

Python, Django, React, REST API, PostgreSQL, Statistical modeling

🚦 06. ETA Prediction & Route Optimization

Problem

Navigation systems only used current traffic conditions, requiring frequent re-routing and unstable arrival predictions.

Solution

We developed a model combining historical congestion patterns with real-time event data for future-aware route planning.

Impact / Result

  • 🕒 Stable arrival time forecasts
  • 🛣️ Reduced need for mid-route changes
  • 🏆 Backed by a granted Korean patent for improving estimated passing speed

Tech Stack

Python, PyTorch, Graph-based modeling, Geospatial analysis

How We Work