📈 Building a Time Series Forecasting ML App During My Internship: A Real-World AI Experience

Hi everyone! 👋

I’m Manognya Lokesh Reddy, a Master’s student in Artificial Intelligence at the University of Michigan-Dearborn. In this post, I’ll be sharing one of the most rewarding experiences of my career so far—building a Time Series Forecasting application during my internship at Donyati India Pvt Ltd. This project helped me understand how AI works in real business settings, and I hope it gives you a glimpse into what it’s like applying ML beyond the classroom.

🧩 The Business Problem
Forecasting plays a critical role in business decisions—whether it's predicting sales, managing inventory, or optimizing resources. The goal of our project was to create an end-to-end Time Series Forecasting application that could accurately predict future trends for multinational clients, replacing outdated manual methods.

⚙️ Tech Stack & Tools
Here’s what I used to build the application:

Programming: Python

Libraries: Pandas, NumPy, Matplotlib, Keras (LSTM)

Modeling: Long Short-Term Memory (LSTM) Neural Networks

DevOps: Docker (for containerization), Git & GitHub

Deployment: AWS (for hosting the application)

🏗️ My Contributions
During my 1-year internship, I worked on:

🔄 End-to-End ML Pipeline
Created a complete ML pipeline—from data ingestion and preprocessing to model training and deployment.

Implemented automated pipelines that reduced manual data processing by 40%.

🧠 Model Optimization
Used LSTM networks to capture temporal dependencies in the dataset.

Tuned hyperparameters and performed feature engineering to improve accuracy.

Achieved a 15% improvement in prediction accuracy over previous models.

🐳 Containerization with Docker
Packaged the entire application into Docker containers to ensure reproducibility and portability.

Result: Deployment setup time was cut down by 50%.

📊 Impact Across Teams
The forecasting tool was deployed across 20+ teams, improving their operational efficiency by 25%.

📚 Key Learnings
Real-world data is messy and inconsistent. Cleaning and preprocessing are just as important as model building.

LSTM models are excellent for capturing time-based trends—but require careful tuning.

DevOps skills like Docker and cloud deployment are essential for taking ML models into production.

It’s not just about building a model—it’s about solving a real problem that saves time and adds value.

🚀 Final Thoughts
This project gave me hands-on experience with both the technical and practical aspects of machine learning in business. It taught me that good ML is not just about the algorithm—it’s about creating systems that are scalable, efficient, and useful.

If you’re a student or early-career professional, I encourage you to:

Work on real datasets

Think about deployment, not just accuracy

Build projects that solve actual business problems