How nipunbatra/stt-ai-teaching works
1. What this repository is
- This is a course-materials monorepo for CS 203, not a runtime application.
- Its main job is to author, generate, and publish slides, labs, demos, diagrams, and reference artifacts for a semester-long AI software engineering course.
2. Primary boundary: source content vs generated output
- The repo cleanly separates authoring sources from rendered deliverables.
- Source lives in places like slides/, labs/, lecture-demos/, diagrams/, diagram-generators/, datasets/, and prompts/.
- Generated or published artifacts live in html/, pdf/, backup/, theme-previews/, and some checked-in notebook HTML outputs.
3. Course website / publication layer
- The top-level site is driven by Quarto: index.qmd, _quarto.yml, themes.qmd, site_libs/, and the GitHub Actions workflow .github/workflows/quarto-publish.yml.
- This layer produces the public course website and ties together the various lecture/lab assets into a browsable static site.
- Hosting appears to be via GitHub Pages or a similar static publish flow.
4. Slides subsystem
- Weekly lecture slides are authored as markdown in slides/.
- Rendered slide outputs are committed under html/ and pdf/.
- The repo also keeps theme-related artifacts such as custom.scss, slides/iitgn-modern.css, theme-previews/, and build helpers like build-all-themes.sh.
- Slide assets include figures and generated images under slides/images/ and slides/audio/.
5. Lab subsystem
- Labs are delivered as notebooks and rendered HTML in labs/.
- Each week’s lab is a self-contained teaching artifact, usually pairing hands-on code with a pre-rendered HTML version for viewing.
- This subsystem is distinct from slides: labs are interactive/experimental, while slides are presentation-oriented.
6. Lecture demos / executable examples
- lecture-demos/ holds week-specific runnable scripts, notebooks, and small app examples.
- These demos illustrate APIs, validation, labeling, augmentation, Docker, web apps, drift, profiling, and CI/CD concepts.
- They often depend on local sample data in subfolders like lecture-demos/week02/data/ or generated model files such as the Docker demo’s spam_model.pkl.
7. Diagram generation pipeline
- There are two related diagram areas: diagram-generators/ and diagrams/.
- diagram-generators/ contains standalone Python scripts and some .d2 / .bob files for producing explanatory figures.
- diagrams/weekXX/ contains more curated week-specific drawing scripts, and diagrams/figures/ stores outputs.
- These diagrams feed the teaching materials rather than any application runtime.
8. Dataset and annotation assets
- datasets/week03/ and data/ provide example data used in labs and demos.
- The week 3 dataset subtree includes subdomains like image classification, object detection, text classification, and NER, along with Label Studio configs and JSON manifests.
- data/movies.db looks like a local teaching database used in examples or exercises.
- This is a support layer for data-centric teaching workflows, not a production datastore.
9. Assignments and guided exercises
- assignments/ and lab01/ contain course exercises and teaching guides.
- These are small pedagogical units designed to reinforce the weekly topics.
- They sit alongside the larger slide/lab/demo materials rather than forming a separate application.
10. Archive and preservation layer
- backup/ preserves older rendered slide versions in HTML, PDF, and source markdown form.
- outputs/ and some checked-in notebook outputs similarly store generated results for reference.
- This suggests a “publish, then archive” workflow where historical versions are retained for continuity.
11. Automation and build orchestration
- Makefile is the main local build entry point.
- build-all-themes.sh and generate-theme-previews.sh support bulk rendering and theme comparison.
- package.json / package-lock.json and the checked-in site_libs/ indicate a front-end/rendering toolchain is also part of the build environment.
12. External tool ecosystem reflected in the content
- The teaching materials reference and demonstrate tools such as Quarto/Reveal.js, Marp-style slide rendering, GitHub Actions, Docker, FastAPI, Streamlit, Gradio, Label Studio, and various ML libraries.
- These tools matter architecturally because they explain why the repo has multiple output formats and many generated assets.
- The repo is organized around teaching workflows that move from raw examples to rendered course deliverables.
13. End-to-end data flow
- Typical flow: source markdown/notebooks/scripts/data → diagrams and generated figures → rendered slides/labs/demo pages → published static site.
- The course website aggregates the outputs, while the generated artifacts remain versioned in the repository for reproducibility and archival access.
14. How to diagram it
- A useful architecture diagram would center on three source domains: content authoring, asset/data generation, and publication.
- Show slides/, labs/, lecture-demos/, diagram-generators/, and datasets/ feeding into html/, pdf/, and the Quarto-published site, with backup/ as an archival sink.
Main entry points: Makefile Build tools · index.qmd Course site · week09-git-lecture.md Slides source · week10-web-apps-lecture.html HTML output
Components 20
Authoring
Source teaching materials and course content
- Slides source · Weekly lecture slide source files
slides/week09-git-lecture.md - Labs · Weekly lab notebooks and rendered lab pages
labs/week01-data-collection-lab.ipynb - Lecture demos · Runnable scripts, notebooks, and small app examples for lectures
lecture-demos/week10/web-apps/4-fastapi/app.py - Assignments · Course assignments and guided exercises
assignments/week1-assignment.md - Lab guide · Standalone lab and teaching guide materials
lab01/cpcb_api_teaching_guide.md
Generation
Scripts and assets that create figures and rendered artifacts
- Theme assets · Shared slide styling, previews, and theme helpers
slides/iitgn-modern.css - Slide media · Generated images and audio used by slides
slides/images/week10/generate_concept_drift_cases.py - Diagram generators · Standalone scripts and diagram sources for explanatory visuals
diagram-generators/llm_pipeline.py - Week diagrams · Week-specific diagram sources and shared figure outputs
diagrams/week03/labeling_workflow.py
Publication
Static site, rendered outputs, and archived builds
- Course site · Top-level Quarto publication for the course website
index.qmd - Quarto config · Site-wide Quarto settings and navigation
_quarto.yml - Publish workflow · GitHub Actions pipeline that builds and publishes the site
.github/workflows/quarto-publish.yml - HTML output · Committed rendered lecture and lab HTML artifacts
html/week10-web-apps-lecture.html - PDF output · Committed rendered lecture slide PDFs
pdf/week10-web-apps-lecture.pdf - Archive · Archived prior rendered versions of slides and outputs
backup/html/week14-rag-vector-db-lecture.html - Site libs · Checked-in Quarto and Reveal.js runtime assets
site_libs/revealjs/dist/reveal.js
Support
Shared data, prompts, and build orchestration
- Datasets · Prepared datasets and annotation manifests used in labs and demos
datasets/week03/create_datasets.py - Local data · Shared local data used by teaching examples
data/movies.db - Prompts · Reusable prompt text for AI and concept explanations
prompts/docker_concepts.txt - Build tools · Local build entry points and theme generation helpers
Makefile
Connections 22
- Build tools → Slides source: build · Renders lecture sources into site and slide outputs
- Build tools → Labs: build · Renders lab notebooks and pages
- Build tools → Diagram generators: generate · Runs figure and diagram generation scripts
- Build tools → Week diagrams: generate · Builds week-specific diagrams and shared figures
- Build tools → HTML output: render · Produces checked-in HTML deliverables
- Build tools → PDF output: render · Produces checked-in PDF deliverables
- Slides source → Theme assets: uses · Slides rely on theme styling and preview artifacts
- Slides source → Slide media: uses · Slides embed generated figures and audio
- Diagram generators → Slide media: produces · Diagram scripts generate figures used across the course
- Week diagrams → Slide media: produces · Curated diagram scripts contribute shared visual assets
- Datasets → Labs: supports · Prepared datasets back the lab notebooks
- Local data → Lecture demos: supports · Local sample data and databases power demos
- Prompts → Course site: informs · Prompt artifacts are part of the course content exposed on the site
- Site libs → Course site: enables · Quarto and Reveal.js assets support site rendering
- Quarto config → Course site: configures · Global site configuration drives publication behavior
- Publish workflow → Course site: publishes · CI workflow builds and deploys the site
- HTML output → Course site: served by · Rendered HTML artifacts are integrated into the course website
- PDF output → Course site: linked by · PDF slide artifacts are exposed through the site
- Archive → HTML output: archives · Preserves prior HTML builds
- Archive → PDF output: archives · Preserves prior PDF builds
- Assignments → Course site: published on · Assignments are surfaced through the course site
- Lab guide → Labs: extends · Standalone lab guide complements notebook-based labs