About RAMS — Your Personal AI Agent

💡 Inspiration

Students today have to manage much more than just academics. Between assignments, deadlines, exams, hackathons, internships, competitions, and personal goals, keeping everything organized can become overwhelming.

At the same time, most AI assistants are primarily designed to answer questions. They can provide information, but they don't necessarily help users manage their day-to-day activities or take meaningful action.

This inspired us to build RAMS, a personal AI agent designed to go beyond conversations and become a practical assistant for students.

The idea was simple: What if students had one assistant that could understand their needs, remember useful information, manage their tasks, plan their schedules, discover opportunities, and help them decide what to focus on?

🚀 What We Built

RAMS is a student-focused AI agent that combines conversational intelligence with persistent memory and specialized tools.

Its key capabilities include:

AI Conversations: Interact naturally with RAMS using a language model-powered agent.
Persistent Memory: Remember useful information across conversations and use previous context to make interactions more meaningful.
Task Management: Create, update, track, and complete tasks using natural-language instructions.
Intelligent Weekly Planner: Generate study schedules based on tasks, deadlines, available time, and existing commitments. Larger tasks can be divided into manageable sessions.
Hackathon Discovery: Discover individual hackathon listings from Unstop and Devpost and extract important information such as deadlines and eligibility requirements.
Eligibility Evaluation: Evaluate saved opportunities against a student's profile.
Personalized Recommendations: Rank opportunities based on eligibility, interests, skills, deadline feasibility, learning value, and workload compatibility.

Instead of simply providing suggestions, RAMS uses its tools and database to perform useful operations and maintain continuity across interactions.

🛠️ How We Built It

We built RAMS using a modular architecture, separating its different capabilities into specialized services.

The application uses Python and Streamlit for its interface, SQLite for persistent storage, and language model integrations for its conversational intelligence.

Its architecture includes:

An agent and tool-calling system for executing supported operations.
A memory system for storing and retrieving conversational context.
Dedicated task management and availability services.
A weekly planning engine that considers actual tasks and commitments.
An opportunity discovery and extraction pipeline.
Separate eligibility evaluation and personalized ranking services.

RAMS currently supports Google Gemini and includes NVIDIA Nemotron integration through NVIDIA's API.

We are also integrating NVIDIA OpenShell to introduce controlled sandbox execution for agent operations. The integration and automated tests are in place, while live sandbox verification remains ongoing.

📚 What We Learned

Building RAMS taught us that creating a useful AI agent involves much more than connecting a language model to an interface.

We learned how to:

Design tool-calling workflows that allow an AI agent to perform actual operations.
Implement persistent memory and retrieve relevant context across conversations.
Design database-backed systems for managing tasks, schedules, and opportunities.
Build intelligent scheduling logic around deadlines and time constraints.
Extract structured information from web pages while handling missing and ambiguous data.
Separate eligibility from opportunity availability.
Build ranking systems that consider multiple factors instead of relying on a single score.
Design automated tests for individual services and integrated functionality.
Think about security boundaries when introducing external execution environments.

One of our biggest takeaways was that an AI agent needs reliable systems and well-defined tools behind it. Intelligence alone is not enough.

⚡ Challenges We Faced

One of the biggest challenges was making RAMS behave like an actual agent rather than a chatbot that simply generates responses.

1. Persistent Memory

Maintaining useful context across multiple conversations required a dedicated memory and retrieval system. We had to ensure that information could be stored, updated, and retrieved without creating unnecessary duplication.

2. Intelligent Scheduling

Generating a realistic weekly plan was more complicated than simply assigning tasks to available hours. RAMS needed to account for existing commitments, deadlines, task priorities, session durations, and workload limitations.

3. Opportunity Data Quality

Hackathon websites contain inconsistent information. Deadlines can appear in different formats, and some listings contain missing or ambiguous dates. We worked on extraction, validation, source tracking, and availability detection to avoid presenting expired opportunities as actionable.

4. Personalized Recommendations

Finding opportunities is only the first step. We needed separate eligibility and ranking systems so that recommendations could consider a student's profile, interests, skills, deadlines, and workload.

5. Secure Agent Execution

We also explored NVIDIA OpenShell to introduce a controlled environment for agent operations. Implementing the integration and automated tests was an important step, although actual sandbox execution still requires live verification.

🔮 What's Next?

RAMS currently focuses on student productivity and opportunities, but the larger vision is to build an extensible personal AI agent.

Future improvements include completing live OpenShell sandbox verification, strengthening voice interaction, expanding opportunity discovery, improving personalization, and introducing more useful integrations.

We want RAMS to become an assistant that grows with its user, understands their context, and helps them move from planning to execution.

Our vision is simple: AI shouldn't just help you find answers. It should help you get things done.