At Forsk Coding School, our Data Science students don’t just learn theory — they solve real-world business problems using machine learning, AI, and data-driven solutions.
This year’s capstone batch worked on diverse, high-impact projects across industries — from energy trading and manufacturing to healthcare and logistics. Each project reflects not just technical skill, but also strategic thinking, teamwork, and the ability to translate data into action.
Here’s a look at the standout capstone projects from our recent Data Science batch.
Smarter LNG Trading with AI: A Bootcamp Team’s Forecasting Solution
Students: Juan Zhang, Daniel Florit, Jianhui Xu
In the dynamic world of global energy trading, data-driven insights can determine success or failure. That’s the challenge faced by BKW, a Swiss energy company navigating complex LNG (liquefied natural gas) markets.
To address this, a team of Forsk Bootcamp graduates — Juan (MBA), Daniel (Finance), and Jensen (Physicist) — developed an AI-powered forecasting tool that simulates profitability and forecasts LNG routes.
The Problem
LNG traders need to analyze fast-changing conditions like price volatility, shifting demand, and complex cost structures. Manual profitability calculations are slow and prone to error, especially in volatile markets.
The Solution
The team built a deep learning forecasting model using LSTM to predict LNG benchmark prices (TTF, PVB, Henry Hub) for 30-, 60-, and 90-day periods. Their model accounted for unpredictable global events through optimized lookback windows and engineered features.
Forecasts were validated using MAE and adapted for practical accuracy.
At Forsk Coding School, Jaipur, students learn to build and evaluate time-series forecasting models for real-world datasets.
To make results user-friendly, they built a Streamlit web application allowing traders to:
- Simulate voyage profitability
- Compare trade routes interactively
- Adjust regas fees or fuel losses
- Explore visual trade scenarios using maps
The next phase involves integrating real-time price feeds and ensemble models for even stronger performance. This project is a perfect example of how AI and finance meet to revolutionize energy market intelligence.
Real-World Clinical Data Dashboard: DataInspector by TriNetX
Students: Dr. Daniel Rodriguez Gutierrez, Dr. Cemil Kerimoglu
The healthcare industry generates massive datasets daily, but analyzing and sharing insights securely remains a challenge. TriNetX, a global leader in real-world data, offers access to diverse clinical datasets — yet project managers often struggle with evaluating dataset quality and communicating results effectively.
The Vision
Daniel and Cemil built DataInspector, a user-friendly, privacy-aware dashboard that transforms how project managers interact with TriNetX’s clinical data.
The Approach
Their design focused on three key needs:
- Assessing complex datasets efficiently
- Enabling secure, collaborative communication
- Turning raw data into clear visual insights
The Result
DataInspector provides:
- Customizable charts and metrics tailored to user needs
- Privacy-first design ensuring secure information handling
- Exportable visuals and summaries for collaboration
With DataInspector, project managers can explore, filter, and share healthcare data meaningfully — turning complexity into clarity.
This project demonstrates how thoughtful design and data science can make a real impact in healthcare analytics and research workflows.
Anomoldy: AI-Based Anomaly Detection for Plastic Molding
Students: Marcia Cabral, Melvin John, Leon Siegel
Hadi-Plast, a family-owned manufacturer of precision plastic parts since 1977, wanted to automate quality control for faster and more consistent inspection.
The Challenge
Visual defect inspection in plastic manufacturing is often manual, slow, and inconsistent. Each defect — a scratch, missing pin, or corner cut — can affect performance but is hard to detect at scale.
The Solution
The team developed Anomoldy, an image-based anomaly detection system using unsupervised deep learning. Instead of labeling every possible defect, the model learns what “normal” looks like and flags deviations automatically.
They implemented models from Anomalib (Patchcore, PaDiM, EfficientAD) and built an interactive Streamlit app for manufacturers to upload part images, view heatmaps, and detect potential anomalies instantly.
By simulating defects and applying preprocessing like rotation- and scale-aware matching, they achieved 96% accuracy, proving strong generalization even on unseen samples.
This project shows how AI-driven quality control can save time, reduce waste, and bring Industry 4.0 efficiency to traditional manufacturing.
Automating End-of-Site Reports for Construction Projects
Students: Sujay Ray, Amos Schtalheim, PhD, Aiyham Katranji, Marco Taglione
VSL International, a global construction company, relies on detailed End-of-Site (EoS) reports to summarize each project’s performance. Traditionally, these reports take days to prepare, involving manual review of interviews, contracts, and documentation.
The Solution
The student team built an AI-powered, semi-automated system that generates structured EoS reports using natural language processing and GPT-based models.
Their system includes:
- Smart Upload & Translation: Automatically detects and translates documents into English while preserving speaker roles.
- Intelligent Parsing: Organizes transcripts into clear question-and-answer formats.
- Automated Question Matching: Uses GPT models to align interview answers with EoS templates.
- Structured Data Extraction: Captures key fields (Yes/No, KPIs, comments) using prompt engineering.
- Traceable Outputs: Generates audit-ready summaries and CSV tables linked to original content.
Results
- 80% automation of manual reporting
- Faster, more consistent results
- Human-in-the-loop design for expert validation
This AI-driven system turns days of reporting into hours, freeing engineers to focus on insight rather than paperwork.
WaitNoMore: Predicting Waiting Times for Roadside Assistance
Students: David Fritsch, Gaurav Jauhri, and Guilherme Samora
Assistance Partner GmbH, Germany’s second-largest roadside assistance provider, wanted to predict ETA (estimated time of arrival) more accurately for customers needing help.
The Project
Using 17,000+ past incidents, the team tested various machine learning models — from regression to neural networks — to predict waiting times based on data like:
- Geographic coordinates
- Traffic and weather conditions
- Provider workload and vehicle type
They even engineered 65+ additional features for deeper insights.
The Outcome
The analysis revealed that while machine learning models performed similarly to simple averages, the real opportunity lay in process improvements.
At Forsk Coding School, Jaipur, students learn to analyze model performance metrics such as Mean Absolute Error (MAE) to evaluate prediction reliability.
Key recommendations included:
- Adjusting default ETAs based on each provider’s historical performance
- Introducing ETA accuracy as a KPI
- Enhancing data collection with real-time signals
This evidence-based approach helped the company focus on immediate operational gains while preparing for smarter AI models in the future.
Estimating Motorcycle Production Across Europe
Students: Nargiz Rüter, Valentina Pavlovic, Giorgio Semadeni
Power Systems Research (PSR), a leader in powertrain analytics, partnered with our team to address a major data gap — inconsistent motorcycle production data across Europe.
The Approach
The students engineered a data pipeline combining registration, import, and export data from Italy, France, Germany, Spain, and the UK.
Each dataset presented unique challenges, from PDFs and images to inconsistent formats. To solve this, the team used a hybrid extraction system combining:
- Rule-based logic for structured data
- OCR for image-based text
- LLM-assisted parsing for complex sources
Using the formula Production = Registrations + Exports – Imports, they inferred national production volumes with 95% confidence intervals to quantify uncertainty.
Results
- Unified, standardized registration datasets for all five countries
- Estimated production shortfall of ~1.99 million units
- Automation strategies for continuous data enrichment
This scalable data pipeline provides PSR with actionable market intelligence, transforming scattered public data into business-ready insights.
Final Thoughts: Data Science in Action
These capstone projects show the power of hands-on learning. Each team tackled real business challenges, applying AI and machine learning to create meaningful, measurable results.
From forecasting global energy prices and automating construction reports to improving healthcare analytics and quality control, these projects highlight the diversity of what data science can achieve.
At Forsk Coding School Jaipur, we believe the best way to learn data science is by doing. Our students graduate with not just knowledge — but experience, confidence, and a portfolio of real-world impact.
Ready to start your journey in Data Science?
Join our next batch and turn your ideas into innovation.
📍 Forsk Coding School – Jaipur
Empowering the next generation of AI and Data Science professionals.
