prince-chhirolya/projects-series

This repository is a portfolio-style monorepo of many standalone ML/data-science projects, not a single integrated application. The main architectural boundary is folder-per-project: each directory contains its own notebook, sample data, and sometimes a small app or model artifact.

1. Top-level shape

- The root contains only repository hygiene files plus dozens of independent project folders.

- There is no shared backend, central package, or common runtime across all projects.

- .github/dependabot.yml suggests dependency maintenance is handled at the repo level, but the projects themselves are isolated.

2. Common project pattern

- Most folders follow the same structure: README.md + one or more notebooks (.ipynb) + local datasets (.csv, images, JSON, videos) + occasional trained outputs.

- This means the primary data flow is usually: local dataset → notebook/script → trained model / analysis output → optional demo UI.

3. Notebook-centric analysis and experimentation

- Many projects are pure notebooks for exploratory analysis, model training, and evaluation.

- Examples include statsmodels analysis, causal inference, epidemic modeling, Lotka-Volterra simulations, image segmentation, GAN augmentation, and several classification notebooks.

- These are typically self-contained and use the notebook as both implementation and documentation layer.

4. Classical ML / tabular prediction track

- A large cluster focuses on tabular ML with CSV inputs and standard supervised learning workflows.

- Examples: diabetes prediction, traffic volume prediction, cancer prediction, frog toxicity prediction, used-car price prediction, fake news detection, writer fingerprinting, Naive Bayes classification.

- The architectural pattern is: CSV data → preprocessing → model training/evaluation → saved model or notebook output.

5. Deep learning / vision track

- Another cluster uses CNNs, ViT, YOLO, OpenCV, and related computer vision tooling.

- Examples: road sign recognition, cats vs dogs classification, anomaly detection in medical images, object detection web app, safety helmet detection, hand sign detection, multiple hand detection, image forgery detection, vision transformer image classification.

- These projects often include heavier artifacts such as pretrained weights (yolov8n.pt), trained model folders, or dataset builders.

6. Interactive web app layer

- Some projects add a small UI layer on top of model or data logic.

- Streamlit appears in multiple places: the data-centric web app, the YOLOv8 object detection app, and the population modeling web-sharing examples.

- One project uses a Flask-style app structure for image forgery detection (app.py, templates, static, uploads), while another uses Gradio for image colorization.

- In these cases the flow is: user input → model/analysis logic → rendered result in web UI.

7. Data visualization and geospatial reporting

- Several projects are centered on visual analytics rather than prediction.

- Examples: Folium crash maps, Plotly/Mapbox geospatial visualization, Streamlit-based map/plot explorer, stats visualizations in notebooks.

- These projects typically transform CSV data into charts or maps, with no persistent service layer.

8. NLP / text / audio projects

- There is a distinct set of language and speech tasks: chatbot with NLTK/TensorFlow, email spam classification with LSTM and Word2Vec, auto-tagging content categorization, fake news detection, knowledge graph extraction, and Whisper speech recognition fine-tuning.

- The data flow here is mostly text corpus or intents JSON → preprocessing/tokenization/embeddings → model or extracted structure.

9. Simulation and scientific modeling

- The population modeling folder contains multiple subdomains: epidemic SIR variants, Game of Life, and Lotka-Volterra dynamics.

- These are simulation-heavy and often include interactive notebooks, plots, phase-space analysis, and parameter fitting.

- The “Web_Sharing” subfolder shows these simulations can be exported to lightweight HTML or Streamlit/Datapane views.

10. Artifacts and pretrained outputs

- Some folders include saved assets that are part of the project boundary: model.json, trained_model/, model.onnx, yolov8n.pt, dataset metadata, HTML examples, and generated media files.

- These artifacts are not shared globally; they belong to the specific project that produced them.

11. Storage and I/O model

- The repo is file-based and local-first.

- Inputs are usually bundled alongside code as CSVs, images, videos, or JSON.

- Outputs are likewise local: notebooks, exported HTML, trained weights, or app uploads folders.

- There is no indication of a central database, message bus, or cloud service dependency.

12. Architecture takeaway for a diagram

- The most accurate diagram is not a single system diagram, but a hub-and-spoke collection of independent mini-systems.

- At the top level: “Personal Projects Series” as the repository container.

- Below that: several parallel lanes such as tabular ML, computer vision, NLP/audio, simulation, and interactive web demos.

- Each lane typically connects data files → notebook/script → optional model artifact/UI, with minimal coupling across folders.