Layer 01
Midterm mini-project
A focused AI application with structured output, validation, tests and documentation.
Student Projects
Build an AI application and a document assistant.
Build a midterm AI application and a document-based capstone with retrieval, cited answers, evaluation and controlled tool use.
- Midterm launches in Week 5
- Final capstone by Week 10
- Portfolio-focused outputs
- Built to be explained in interviews
Layer 02
Final capstone
A richer document-based AI system with ingestion, retrieval, grounded generation, evaluation, and one controlled tool-calling workflow.
Midterm Mini-Project
First, students build a focused AI application.
The midterm launches at the end of Week 5 and gives students an early chance to build a complete AI application before embeddings, vector databases, RAG, and agents enter the picture.
Designed To Test
- Define a user problem clearly.
- Build a focused AI workflow.
- Validate outputs and handle failure cases.
- Explain limitations instead of hiding them.
Example Use Case
A resume and job description analyzer that extracts requested skills, identifies matching and missing skills, generates role-specific interview questions, and returns the result in a controlled structured format.
Minimum Deliverables
- A clearly defined problem and intended user.
- At least one machine-learning or LLM capability.
- Structured output with validation and basic error handling.
- At least 10 representative test examples, including failure cases.
- A simple interface such as Streamlit, Gradio, or an equivalent CLI.
- A readable repository with a concise README.
- A short live demonstration and explanation of limitations.
Final Capstone
Then students build a document-based AI assistant with retrieval and evaluation.
Work with a source document collection and generate answers supported by citations.
Students combine ingestion, retrieval, grounded generation, evaluation, and one controlled tool-calling workflow into one application.
Minimum Functional Requirements
- A clearly defined user group, problem statement, and success criteria.
- At least five source documents with visible metadata.
- Document extraction, cleaning, chunking, and repeatable ingestion.
- Embedding generation and vector-database storage.
- Semantic retrieval with configurable top-k results.
- Grounded answer generation with citations.
- Explicit handling for unsupported questions.
- A basic user interface.
- At least 20 evaluation questions, including unsupported and adversarial questions.
- One controlled agent capability with 2 to 3 approved tools.
- Visible tool calls and human approval before consequential actions.
- Logging, documentation, architecture explanation, and reproducible startup instructions.
Example Capstone Themes
- College Information Assistant
- Placement Preparation Assistant
- Research Paper Assistant
- Corporate Policy Assistant
- Product Support Assistant
Controlled Agent Extension
Add a workflow that uses approved tools and requires human approval before consequential actions.
- Search documents, compare skills, and generate a learning plan.
- Retrieve a policy, calculate entitlement, and draft a human-review request.
- Retrieve research claims and create a comparison summary.
What Students Finish With
The output is more than a single demo.
- Machine-learning notebook and evaluation report.
- Midterm AI text-analysis application.
- Semantic-search implementation.
- Vector-database ingestion and query scripts.
- Complete RAG and agent capstone repository.
- Architecture and data-flow diagrams.
- Evaluation dataset and summarized results.
- Demo script or short demonstration asset.
- Resume-ready project description and contribution summary.
Resume-Style Proof Point
Built a document-based AI assistant using Python, embeddings, a vector database, and retrieval-augmented generation with cited answers, evaluation across 20-plus test questions, and a controlled tool-calling workflow.
Midterm Evaluation Areas
- Problem definition and user value.
- Technical implementation.
- Testing, evaluation, and failure analysis.
- Code quality and documentation.
- Demonstration and technical explanation.
Capstone Evaluation Areas
- Problem definition and user value.
- Document ingestion and data quality.
- Retrieval design and quality.
- Answer quality, grounding, and citations.
- Evaluation methodology and failure analysis.
- Agent design, tool safety, and stopping behavior.
- Application usability and error handling.
- Code quality, documentation, and reproducibility.
- Demonstration and technical explanation.
Next Step
If you want to build projects that go beyond AI demos, start here.
KalyrAITech is designed for students who want structured, portfolio-relevant project work across machine learning, LLM applications, retrieval, RAG, and controlled tool use.