Building an AI Powered WhatsApp Message Notification Router in the 24 Hour HackerRank Orchestrate Challenge

Hackathons are more than competitions. They are opportunities to learn under pressure, experiment with new ideas, and build practical solutions within strict time constraints.

Recently, I participated in the 24 Hour HackerRank Orchestrate Challenge, where I developed an AI Powered WhatsApp Message Notification Router, a system designed to intelligently decide whether an incoming message should immediately notify the user, be added to a digest for later viewing, or be muted.

The Problem

Every day we receive hundreds of notifications from personal chats, work groups, promotional messages, OTPs, business updates, and spam. Most notification systems treat every message the same, leading to unnecessary distractions and notification fatigue.

The challenge was to build a smarter notification system that understands the context of a message instead of relying on simple keyword matching.

My Approach

I designed a hybrid AI pipeline that combines semantic understanding, machine learning, deterministic rules, and personalization.

The workflow looks like this:

1. Load multiple datasets including messages, user preferences, business accounts, groups, and historical interactions.
2. Clean and preprocess every incoming message.
3. Generate semantic embeddings using the Sentence Transformers all MiniLM L6 v2 model.
4. Extract handcrafted features such as urgency, payment related terms, promotions, spam indicators, message length, and conversation type.
5. Train Logistic Regression models to predict both the message category and the appropriate notification action.
6. Apply a rule engine that overrides predictions for safety critical scenarios including scams, phishing attempts, OTP requests, and highly urgent messages.
7. Personalize the final decision using user behavior, muted groups, trusted contacts, and previous interaction history.
8. Retrieve supporting historical evidence, calculate confidence scores, and generate explainable reasoning.
9. Export the results into an output.csv file along with a detailed execution log for transparency and debugging.

The final output contains:

• Message ID

• Notification Action

• Message Type

• Confidence Score

• Reasoning

• Supporting Evidence IDs

Technologies Used

Python

Sentence Transformers

all MiniLM L6 v2

Scikit Learn

Logistic Regression

Pandas

NumPy

Feature Engineering

Rule Based AI

Semantic Embeddings

Explainable AI

What I Learned

This project taught me far more than simply training a machine learning model.

I learned that real world AI systems rarely depend on a single algorithm. Instead, they combine multiple approaches to improve accuracy, reliability, and explainability.

Some of my biggest takeaways were:

• Building semantic embeddings for better language understanding.
• Designing meaningful handcrafted features that improve prediction quality.
• Combining machine learning with deterministic business rules.
• Understanding why explainability is essential for AI systems.
• Personalizing AI decisions based on user behavior rather than creating one generic solution.
• Structuring a modular Python project that is easier to debug and maintain.
• Creating transparent logs and evidence based outputs for auditing predictions.

Challenges I Faced

The biggest challenge was the time limit.

Building an end to end AI pipeline in just 24 hours required balancing speed with quality.

Some of the obstacles included:

• Understanding the complete problem statement quickly.
• Designing meaningful features from limited data.
• Deciding when machine learning predictions should be overridden by hard rules.
• Handling edge cases such as phishing, scams, OTP messages, and urgent notifications.
• Integrating personalization without making the architecture overly complex.
• Generating explainable outputs instead of simple predictions.
• Debugging multiple modules while keeping the project modular and organized.

Every challenge pushed me to think more like an engineer than just a programmer.

Looking Back

One of the most rewarding aspects of this project was realizing that building AI is not only about achieving high prediction accuracy. It is about creating systems that users can trust.

By combining semantic understanding, supervised learning, feature engineering, rule based reasoning, and personalization, I built a notification routing system that is intelligent, explainable, scalable, and user focused.

Although there is always room for improvement, such as incorporating multimodal models, online learning, or reinforcement learning for continuous personalization, this project strengthened my understanding of how production style AI systems are designed.

Final Thoughts

The HackerRank Orchestrate Challenge was an incredible learning experience that pushed me outside my comfort zone. Working under a 24 hour deadline taught me how to prioritize, solve problems efficiently, and build practical AI solutions with real world impact.

Projects like this remind me that every challenge is an opportunity to learn something new, improve my engineering skills, and move one step closer to building AI systems that genuinely make everyday technology smarter.

Thank you, HackerRank, for creating such an exciting challenge. I am looking forward to applying these learnings in future AI and data science projects.