How oam11/automl works
File structure
• Top-level files:
• app.py – streamlit application entrypoint (defines UI, routing, session state).
• index.html – custom static page (e.g., landing page or template override).
• requirements.txt – dependency list.
• .devcontainer/ – VS Code dev container config for a reproducible dev environment.
• README.md, LICENSE – documentation.
• There are no separate “frontend” and “backend” folders; the Python process hosts both. Logic is modularized within app.py (or imported modules).
Extract architectural insights from the README
• UI layer: Streamlit sidebar + pages for data upload, EDA, ML workflows, and an “Ask Questions” chat interface.
• Business Logic:
• Data Management (upload, encoding detection, persistence via Streamlit session state or temporary storage).
• EDA Module (ydata-profiling report generation, visual exploration).
• AutoML Module (classification, regression, clustering, anomaly detection, time series) via PyCaret.
• NLP Interface (LangChain + Groq’s Llama 3.3 for context-aware QA).
• External dependencies & services: Groq API for LLM inference.
• Output artifacts: downloadable trained model files, profile reports, cluster assignments.
1. User Browser (Client)
2. Streamlit Server (Python Runtime)
• UI Layer (Streamlit widgets, pages)
• Data Manager
• EDA Engine (ydata-profiling)
• AutoML Engine (PyCaret submodules)
• NLP Engine (LangChain orchestrator + Groq LLM calls)
3. Data Storage
• In-memory / session state (Streamlit cache)
• Temporary file store (uploaded CSVs, sample datasets)
4. External Service
• Groq API (Llama inference)
• Browser → Streamlit UI: user actions (file upload, button clicks, chat queries)
• Streamlit UI → Data Manager: store/retrieve datasets in session state or disk
• Data Manager → EDA Engine: raw data passed for profiling
• EDA Engine → Streamlit UI: report objects and plots rendered
• Streamlit UI → AutoML Engine: train/evaluate models, retrieve metrics
• AutoML Engine → Streamlit UI: visualizations, downloadable model artifacts
• Streamlit UI → NLP Engine: user questions + data context forwarded
• NLP Engine → Groq API External Service: REST/gRPC calls with prompt/context
• Groq API → NLP Engine: LLM responses → rendered in UI
c. Architectural patterns & design principles:
• Single-process modular design (monolith) using Streamlit’s page routing.
• Separation of concerns: UI vs. data management vs. ML vs. NLP.
• Stateless external inference (Groq API) vs. stateful session data.
• Pluggable ML engines (via PyCaret) and profiling tool (ydata-profiling).
d. Key technologies & libraries (annotate each component):
• Streamlit (UI/control flow)
• PyCaret (AutoML)
• ydata-profiling (exploratory profiling)
• LangChain + groq (natural language interface)
• Pandas/Numpy (data handling)
• Plotly/Matplotlib/Seaborn (visualizations)