Memory with a purpose
Five memory types keep conversations, student context, task state, vectors, and enrollment data together.
Agentic AI / Academic advising
CourSelect
A better path
to your next course.
SYSTEM / COURSELECT01 → 04
01Student goalAcademic context + intent
02RetrievalVectors + eligibility
03AdvisorGemini → local → cloud
04PersistenceCockroachDB + S3 reports
8,983courses in the catalog
242academic departments
4response fallback tiers
An academic advisor that connects a student’s goals, course history, and enrollment rules to a searchable university catalog.
Recommendations need context.
Course discovery is only half the problem. A useful recommendation also needs to respect prerequisites, a student’s major, and their completed academic history.
A deliberate path through the problem.
A Flask application brings course browsing, advising, and enrollment together. Semantic retrieval finds relevant courses; the advisor uses student context and eligibility checks to explain the recommendations.
PythonFlaskCockroachDBGeminiAWS S3
The choices that shape the work.
A response when models fail
The engine falls back from Gemini to Ollama, Mistral Cloud, and finally direct vector results.
Keep the slow work off the path
Parallel pre-fetching, two-level caching, and background S3 uploads reduce work on the response path.
What the project delivers.
The repository documents a complete advising workflow: catalog exploration, contextual recommendations, enrollment, and protected historical records. Its search cache persists across application restarts.
Resilience is part of the product.
The central engineering choice is graceful degradation: the catalog and retrieval layer remain useful even when a model provider is unavailable.