Applied AI
In progressAI Tutor
A tutoring application that uses a language model to guide learners, backed by a structured API, retrieval of relevant material, and an evaluation process for answer quality. Evaluation and quality work is ongoing.
Model evaluation and quality validation are ongoing.
- Status
- In progress
- Type
- Applied AI
- Technology
- Python
- FastAPI
- PostgreSQL
- SQLAlchemy
- LLM evaluation
The challenge
A tutor has to be helpful and accurate at the same time. Responses need to draw on the right material, stay on topic across a conversation, and improve over time without a change in one place quietly making answers worse somewhere else.
The approach
AI Tutor pairs a FastAPI backend with PostgreSQL for application data. Tutoring workflows assemble the relevant context for each exchange, and an evaluation process scores model responses so prompt and workflow changes can be compared rather than guessed at. The evaluation rounds, quality validation, and a backlog of further evaluation work are still in progress.
Engineering highlights
Backend architecture
A FastAPI service with SQLAlchemy models over PostgreSQL, organized so tutoring logic, data access, and model calls stay separate.
AI tutoring workflows
Structured flows guide how the tutor responds, rather than passing raw input straight to a model.
Retrieval and context management
Relevant material and conversation history are selected and supplied to the model for each response.
Model evaluation (V1 and V3)
In progressResponses are evaluated against defined expectations to measure quality. The V1 and V3 evaluation rounds are under way.
Quality validation
In progressAutomated tests and evaluation results are used to validate answer quality and guide improvements to prompts and workflows.
Evaluation backlog
In progressFurther evaluation work is planned and being worked through.
What is still in progress
- Model evaluation (V1 and V3)
- Quality validation
- Evaluation backlog
This work is ongoing. This page will be updated as it is completed; until then the project is described as in progress rather than finished.
Related services
A note on confidentiality
This summary describes the problem and the engineering approach. Source code, data, credentials, and organization-specific details are not shared publicly. We are happy to discuss the technical decisions in more depth in conversation.