Inspiration
As a student at IMSA, I noticed a common hurdle for prospective families: our website is a treasure trove of information, but its depth can be overwhelming. During admissions season, parents often struggle to find specific answers, like exact deadlines or fee structures, especially when navigating via mobile or if English isn't their primary language. I was inspired to create a tool that acts as a "digital bridge," transforming a complex web directory into a simple, natural conversation.
What it does
IMSA Admissions Buddy is a specialized AI assistant that provides instant, factual answers to questions about the IMSA application process. Unlike general search engines, it is grounded in official IMSA data. It can:
- Answer specific queries about deadlines, requirements, and tuition in seconds.
- Communicate fluently in multiple languages, including Spanish and French.
- Filter out irrelevant queries to stay focused strictly on admissions.
- Clarify complex topics, such as the distinction between tuition-free status and residential fees.
How we built it
The project was developed using Chatbase as the core platform, integrated with a custom knowledge base derived from the IMSA admissions portal. The technical brain of the project lies in the System Prompt Engineering. I designed a specific instruction set that defines the bot’s persona, ensures it only uses verified data (preventing hallucinations), and maintains a concise, two-sentence conversational style.
The logic follows the equation: $$Accuracy = \text{(Official Data)} + \text{(System Prompt Guardrails)}$$ I also configured the NLP (Natural Language Processing) settings to handle multilingual inputs, allowing for a more inclusive user experience.
Challenges we ran into
The biggest challenge was hallucination management. Early versions of the bot would try to answer general questions or pull outdated info from the broader web. I had to iteratively refine the system prompts to create a closed-loop environment. Another challenge was the time constraint of the hackathon, which forced me to prioritize core functionality, like accuracy and language support over aesthetic UI features.
Accomplishments that we're proud of
I am particularly proud of the bot's multilingual accessibility. Seeing the bot instantly translate complex Illinois residency requirements into Spanish was a major win for community equity. I am also proud of the guardrail system. The bot successfully identifies and deflects off-topic questions (like pop culture or unrelated news), which is critical for a professional school tool.
What we learned
This project taught me the power of Prompt Engineering. I learned that the performance of an LLM isn't just about the data you give it, but the constraints you set for it. I also gained a deeper understanding of how AI can be used for social good by making educational resources more equitable and easier to navigate for non-native English speakers.
What's next for IMSA Admissions Buddy
The next step is Scalability. I plan to expand the knowledge base to include "Student Life" by feeding the bot transcripts from Zoom orientation sessions. This would allow parents to ask the bot about specific details mentioned in meetings they might have missed. Eventually, I see this as a service that could be licensed to other specialized schools to help manage their admissions traffic more efficiently.
Built With
- chatbase
- knowledge-base-indexing
- llm
- natural-language-processing
- system-prompt-engineering