AdmitPlus.AI

Inspiration

Studying abroad is one of the most life-changing opportunities for students, but the process is fragmented and stressful. • Students struggle with school matching, essay writing, and managing dozens of deadlines. • Agencies lack scalable tools to support large numbers of clients efficiently.

We asked ourselves: What if AI agents could handle these repetitive, high-friction tasks end-to-end, so that humans could focus on real guidance and decision-making? That idea inspired us to create AdmitPlus.ai — an AI-powered study-abroad operating system.

What it does

AdmitPlus.ai is an AI-powered study-abroad operating system that helps both students and agencies. • For students: AI agents match them with best-fit universities, generate essay drafts and outlines, and track application deadlines. • For agencies: A CRM-style dashboard shows all students’ progress, documents, and upcoming deadlines, enabling advisors to manage more clients with less effort. • AI Orchestration: Multiple specialized agents (matcher, essay co-pilot, deadline watchdog) collaborate through a workflow engine, ensuring end-to-end support from school search to acceptance.

How we built it

Our system consists of several coordinated components:

1. AI Agents • School Matcher Agent: Suggests best-fit programs based on academic background, test scores, budget, and preferences. • Essay Co-Pilot Agent: Generates outlines and drafts, then helps refine essays iteratively. • Deadline Watchdog: Monitors upcoming deadlines and alerts students and agencies.
2. Data & Knowledge Layer • Curated a structured database of ~2,000 universities with official names, aliases, and rankings (QS, THE, USNews, ARWU). • Indexed essay examples and admission statistics with a vector search layer for retrieval-augmented generation (RAG).
3. Workflow Orchestration • Built with FastAPI and LangChain/LangGraph to coordinate multiple agents. • Integrated caching and memory to let agents “remember” student context across sessions.
4. AWS Infrastructure • Model hosting on SageMaker. • DynamoDB for structured school/program records. • S3 for student documents (resumes, transcripts, essays). • CloudFront + Amplify for a lightweight web frontend.

Challenges we ran into

• Data Normalization: University names vary widely (e.g., “UCB” vs. “UC Berkeley” vs. “加州大学伯克利分校”). We solved this using an alias mapping system plus vector embeddings. • Essay Generation Quality: LLMs often produce generic or repetitive drafts. We added RAG + critique loops to improve originality and persuasiveness. • Deadline Management: Different programs publish deadlines in inconsistent formats. We built a parser to standardize dates into ISO 8601 and stored them in DynamoDB. • Balance of Automation & Human Control: Users wanted AI assistance but not full automation. We designed the system to let users review, edit, and override agent outputs easily.

Accomplishments that we're proud of

• Built an end-to-end AI workflow in just a hackathon sprint, integrating school matching, essay generation, and deadline reminders. • Created a structured database of 2,000 universities with standardized names, aliases, and global rankings (QS, THE, USNews, ARWU). • Deployed AI agents on AWS infrastructure (SageMaker, DynamoDB, S3), making the system scalable and cloud-ready. • Designed a user-friendly interface for both students and agencies, bridging B2C and B2B needs in one platform. • Early pilot feedback from agencies confirmed that the tool saves significant time compared to their current manual workflows.

What we learned

Throughout this project, we learned how to: • Design AI agents that can work orchestrated workflows instead of isolated tasks. • Combine structured data (university databases, deadlines, requirements) with LLM-based reasoning for personalized recommendations. • Leverage AWS cloud services for scalability and reliability, e.g., deploying models with Amazon SageMaker, and using DynamoDB + S3 for efficient storage. • Evaluate how students and agencies perceive trust, explainability, and usability in AI recommendations.

What's next for AdmitPlus.AI

• Smarter Matching: Incorporate program-level admissions data (GRE/TOEFL ranges, acceptance rates) to improve recommendation accuracy. • Essay Quality Boost: Fine-tune models with domain-specific datasets to produce more original, persuasive essays. • Workflow Automation: Add more AI agents (e.g., document reviewer, visa checklist agent) to expand coverage beyond admissions. • Global Rollout: Launch pilots with overseas education agencies in Asia and the Middle East. • Trust & Transparency: Implement explainable AI outputs, so both students and agencies can see why a recommendation or essay suggestion was made.

Built With

• amazon-web-services
• arwu)
• aws-cloudfront-+-amplify-(frontend-hosting)-?-databases:-dynamodb-(production)
• aws-dynamodb-(structured-data)
• aws-s3-(file-storage)
• ci/cd
• custom-embedding-pipelines-?-developer-tools:-docker
• docker
• dynamodb
• fastapi
• flake8
• github-actions-(ci/cd)
• isort
• langchain
• langgraph
• langgraph-?-cloud-services:-aws-sagemaker-(model-hosting)
• languages:-python
• llms
• mongodb-(prototype)
• mypy)-?-other-apis:-university-ranking-apis/datasets-(qs
• pre-commit-hooks-(black
• python
• rag
• react
• redis
• redis-(caching-&-session-memory)-?-ai-/-ml:-llms-(via-sagemaker-+-api)
• retrieval-augmented-generation-(vector-database)
• s3
• sagemaker
• the
• typescript
• typescript-(for-backend-+-frontend)-?-frameworks:-fastapi
• usnews
• vector