Inspiration

The current solution for academic guidance primarily involves an academic advisor, a faculty advisor, online resources, and annual calendars. While these resources exist, they often don’t fully support students in understanding degree plans or accessing personalized advice. Many students struggle to navigate these services, leading to confusion and wasted time and may miss out on the most convenient academic path.

Our interviews found that most students rely on advice from senior classmates, but this guidance is limited to their own experiences, which may only apply to some. Others feel uncomfortable seeking help from faculty advisors because they are uneasy with the lengthy bureaucratic procedures.

Our findings show more than 2 students from the class of 2024 who didn’t graduate and another from the class of 2025 who won't graduate because they weren't informed enough and took courses that didn’t meet their degree requirements, forcing them to retake classes at their own expense. More than 3 students shared that they had enrolled in a practicum project this fall but later dropped the course due to dissatisfaction, and when they attempted to switch tracks, they realized too late that the deadline for other course registrations had passed. They now have to make compromises they had not planned before the fall semester.

These experiences highlighted the need for a solution that provides clear, personalized, and accessible academic guidance, leading to the creation of the Academic Advisory AI Assistant called Olga. Olga is a mobile app that collects and trains data from students' past experiences, including course reviews, degree plans, course catalogues (Africa and Pittsburgh), the student handbook, faculty course evaluations, class schedules, academic calendars, and more. This is to provide you with 24/7 academic advisory services, accessible at anytime from any location.

What it does

Our mobile app assists students by: Welcome to our user-friendly chat interface designed to provide instant and reliable answers and guidance on degree progress, requirements, and academic paths. You can count on the information you receive to be accurate, current, and drawn from students' past experiences, including comprehensive course reviews, degree plans, course catalogues (Africa and Pittsburgh), the student handbook, faculty course evaluations, class schedules, academic calendars, and more.

It includes getting personalized course recommendations based on your LinkedIn profile, CV, past courses (SIO), degree requirements and the above-mentioned sources. Stay ahead with alerts and notifications for crucial academic deadlines, including registration periods, drop dates, and missing prerequisites.

How we built it

System Architecture and Development Process

The system architecture is built with a three-layer structure and incorporates Natural Language Processing (NLP) capabilities to enhance the chatbot’s understanding of user queries. Below is a step-by-step process of how we built the system, with explanations for each component.

Step 1: Mobile UI Layer Development

The first step was to build the Mobile User Interface (UI) where users can interact with the chatbot. This interface allows students to ask questions about their academic progress, degree requirements, and receive course recommendations.

Key Steps:

We used React Native to build the mobile app for both Android and iOS platforms. React Native's cross-platform capabilities ensured that we could deploy the app efficiently to both operating systems without developing separate apps.
The UI design prioritized user experience, making it easy to ask questions and receive responses in a clear, readable format.

Step 2: Backend Services Layer Setup

The next step involved developing the Backend Services Layer, which handles communication between the chatbot and the various data sources, such as student records and the student handbook.

Key Steps:

We used Node.js for backend logic, as it offers excellent scalability and real-time capabilities, which are crucial for a responsive chatbot.
REST APIs were built to fetch and send student data, course recommendations, and academic progress details. These APIs communicate with the mobile UI to ensure the chatbot responds with relevant information.
The backend connects to the institutional database to retrieve up-to-date academic information.

Step 3: Building the NLP & Rule-Based Chatbot

With the backend and UI in place, we integrated Natural Language Processing (NLP) into the chatbot, allowing it to interpret and respond to student queries conversationally. We also implemented rule-based logic for predefined responses.

Key Steps:

We integrated Dialogflow as our NLP framework to process user input. This allowed the chatbot to understand a wide variety of queries, such as “Check degree progress” or “Get course recommendations.”
The NLP layer could also handle variations in phrasing. For example, the chatbot understood the difference between “What courses should I take?” and “What’s next in my academic plan?”
Rule-Based Logic was used for specific scenarios where predefined academic guidance was needed, such as course prerequisites or degree requirements. This ensured that responses were accurate and aligned with institutional policies.

Step 4: Integrating Data Sources

Once the chatbot was capable of understanding and responding to queries, we focused on integrating it with various data sources to ensure it could provide accurate, real-time information.

Key Steps:

Degree Requirement Data: The chatbot pulls information directly from the student handbook and other resources, ensuring students receive up-to-date and accurate degree requirement information.
Course Recommendations: Based on students’ academic progress, the chatbot suggests courses using predefined logic. This ensures that the recommendations align with degree requirements and student goals.
Alerts and Notifications: The chatbot provides real-time alerts, such as registration deadlines or notifications about missing prerequisites. These notifications are delivered via push notifications to ensure students stay informed.

Step 5: Adding Retrieval-Augmented Generation (RAG)

We incorporated a RAG (Retrieval-Augmented Generation) system to enhance the chatbot’s response accuracy. RAG is a content store with domain knowledge, which allows the chatbot to access specific information beyond what the base LLM knows.

Key Steps:

The RAG system connects to external data sources, ensuring the chatbot retrieves the most accurate and relevant information for user queries.
This setup allows the system to scale, ensuring that the chatbot remains useful as new content is added or updated in the knowledge base.

Step 6: Developing the Assessment LLM Agent

To ensure the chatbot’s accuracy, we implemented an Assessment LLM Agent. This agent continuously assesses the quality of the chatbot’s responses and updates the base model accordingly.

Key Steps:

The Assessment LLM Agent evaluates the chatbot’s performance by comparing its responses to expected outputs. If the response is deemed unsatisfactory, the agent flags it for updates.
This continuous assessment loop ensures that the chatbot improves over time, enhancing the user experience and the accuracy of responses.

Step 7: Testing and Optimization

The final step in the development process was extensive testing to ensure that the system worked smoothly across all components: UI, backend services, NLP capabilities, and data integration.

Key Steps:

We tested the chatbot with various user queries to ensure it could handle different types of input, including misspelled words and complex sentence structures.
We optimized the performance of the system, ensuring that responses were delivered in real time without significant delays.

Step 8: Deploying and Monitoring

Once all components were tested and optimized, we deployed the system. Post-deployment, we continuously monitor the chatbot’s performance to ensure it meets user expectations.

Key Steps:

We set up logging and monitoring tools to track user interactions and performance metrics.
Feedback loops were established to quickly address any issues raised by users, ensuring that the chatbot remains a reliable tool for students.

Conclusion

This step-by-step process shows how we built an efficient and intelligent chatbot system, leveraging the power of NLP and rule-based logic while integrating real-time data sources. By continuously assessing and improving the system, we ensure that students can rely on it for accurate academic advice and guidance.

Challenges We Ran Into

One of the biggest challenges was integrating multiple data sources, such as course catalogs, student records, and degree plans, into a cohesive system that delivers accurate, real-time information. Ensuring the chatbot could handle diverse query phrasing through NLP was another obstacle, especially when parsing more complex academic queries. Additionally, maintaining the responsiveness and scalability of the system while processing large amounts of data was a challenge we had to overcome, particularly in the recommendation engine.

Accomplishments That We're Proud Of

We’re proud of successfully building a fully functional chatbot that delivers personalized academic advice to students in real-time. The integration of NLP capabilities with real-time event notifications has dramatically improved the user experience, allowing students to make informed academic decisions instantly. We are also particularly proud of creating a recommendation engine that tailors course suggestions to each individual student’s academic journey, drawing from multiple data sources like LinkedIn profiles and academic records.

What We Learned

We learned the importance of clean, centralized data management when building systems that pull from multiple data sources. Efficient handling of real-time updates and notifications can greatly enhance user engagement, and the value of sentiment analysis in creating a more empathetic and supportive chatbot cannot be understated. We also gained deeper insights into deploying scalable AI/ML systems and refining our processes for continuous model improvement through feedback loops.

What's Next for OLGA AI Academic Advisor

Moving forward, we plan to expand OLGA’s capabilities by integrating more advanced features such as predictive analytics to forecast student performance and suggest personalized academic plans. We aim to refine the sentiment analysis to make OLGA even more responsive to student needs and emotions. Additionally, we’ll work on integrating more external data sources, such as industry trends and job market requirements, to provide more career-oriented advice alongside academic recommendations.