DataVizhi

• 

  Graph based on Given data

• 

  Tread foreseeing

• 

  Local Ai expert

• 

  Advance concept for data analysi

• 

🏛️ Build Story

Why we started

• We needed local-first analytics that respected data sovereignty and ran fully on CPUs.
• We wanted analysts to get trustworthy insights in minutes without wiring together notebooks, cloud creds, or brittle scripts.

How we built it (milestones)

• Foundation: Stood up ingestion/loader.py for multi-file uploads, then layered validation in utils/data_validator.py so bad data fails early with human-readable errors.
• Schema IQ: Built schema/analyzer.py to fingerprint columns, spot dates and numerics, and hand that context to downstream steps.
• Decision Engine: Added decision_engine/rules.py to auto-suggest the right analysis path (forecasting, anomalies, clustering) based on schema signals.
• Models that fit offline: Implemented ARIMA forecasting, Isolation Forest anomalies, and K-Means clustering in ml_models/, tuned for CPU-only speed and predictable defaults.
• Visual Storytelling: Wrapped Plotly charts in Streamlit, emphasizing responsive layouts and downloadability for quick handoffs.
• Local LLM assist: Integrated Ollama in llm/ to turn numbers into narratives while keeping every token on the user's machine.

Principles we kept

• Privacy-first: nothing leaves the box; defaults avoid external calls.
• Fast enough on a laptop: sampling and chunked reads keep 200K+ rows workable.
• Reproducible: single config.py, .env options, and lean requirements.
• Observable: structured logs plus focused tests (tests/verify_backend.py, clustering/validation specs) to catch regressions early.

Built With

• ai
• api
• ollama
• python