Drop in a spreadsheet, say what you want to know, and get an interactive dashboard.

Hillside AI is a web app I built that turns a CSV or Excel file into a dashboard. It profiles the data, talks through what you want to learn, then builds a page of charts, KPI cards and tables that you can filter, rearrange and edit, by hand or by asking the AI. (The project started under the name ONYX, hence the repository name.)

How it works

1. Pick a goal and attach a file	2. Talk it through	3. Edit any chart
Choose sales, customers, operations or trends, or type your own question.	The AI reads the data's profile, points out what stands out, and asks what matters before it builds anything.	Change the chart type, axes and aggregation, drag and resize widgets, or ask the AI to refine the dashboard.

Features

• Upload CSV, TSV and Excel (.xlsx, .xls, .xlsb) files. Uploads are converted to Parquet for fast reloads.
• Profiling: column types, unique values, missing values, min, max and mean, shown as a summary table.
• Automatic dashboards: a rules engine scores every column and picks KPI cards, bar, line, pie, scatter and histogram charts, tables, and maps when it finds latitude and longitude. It also adds views such as totals by day of week or hour of day, and conversion rates for yes/no columns.
• AI chat that builds and refines dashboards, with streamed responses.
• Choose your model: 13 models from Google Gemini, Anthropic Claude, OpenAI and Groq, 7 of them free tier.
• Filters: date ranges (7, 30 or 90 days, year to date, custom) and field filters, applied to every chart.
• Combine data: join two datasets on matching key columns, or stack files with the same layout.
• Projects and accounts: email and password (JWT) or Google sign-in. Work done as a guest moves to your account when you sign up.
• Export the data to CSV, and save the key insights as a PNG.

Architecture

flowchart LR
  UI["Browser<br/>vanilla JS, Plotly.js, GridStack"] -->|REST + streaming| API["FastAPI"]
  subgraph Backend
    API --> R["Routers<br/>auth, projects, datasets, visuals, models"]
    R --> ING["Ingestion<br/>pandas → Parquet"]
    R --> VIS["Visual engine<br/>column scoring, chart specs"]
    R --> LLM["LLM router"]
  end
  ING --> FS[("File storage")]
  R --> DB[("SQLite<br/>SQLModel")]
  LLM --> P["Gemini · Claude · OpenAI · Groq"]

Loading

The backend does the analysis and sends chart specifications; the frontend draws them with Plotly.js and lays them out on a GridStack grid.

Run it locally

git clone https://github.com/vikwillberg/Onyx-AI.git

cd Onyx-AI && python -m venv .venv

source .venv/bin/activate      # Windows: .venv\Scripts\Activate.ps1

pip install -r backend/requirements.txt

uvicorn app.main:app --app-dir backend --port 8002

Then open http://127.0.0.1:8002. The Datasets/ folder has about 30 public sample datasets to try (sales, pizza orders, subscriptions, web sessions and more).

AI providers (optional)

Create backend/.env with a key for any provider you want to use. Groq and Gemini both have free tiers.

GROQ_API_KEY=your-key        # https://console.groq.com
GEMINI_API_KEY=your-key      # https://aistudio.google.com
CLAUDE_API_KEY=your-key      # https://console.anthropic.com
OPENAI_API_KEY=your-key      # https://platform.openai.com

Variable	Default	Purpose
JWT_SECRET_KEY	change-me	Signs login tokens. Set your own in production.
DATABASE_URL	sqlite:///./app.db	Database connection
STORAGE_DIR	./storage	Where uploaded files are kept
CORS_ORIGINS	["http://localhost:3000"]	Allowed browser origins
GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET	(empty)	Turns on "Sign in with Google"

Deploy

render.yaml describes the service for Render: connect the repository as a Blueprint and it installs the backend, generates a JWT secret and starts the server.

Project layout

backend/app/
├── main.py              # FastAPI app; also serves the frontend
├── routers/             # auth, Google sign-in, projects, datasets, visuals, models
├── services/
│   ├── ingestion.py     # file reading, profiling, merge and stack
│   ├── visuals.py       # column scoring and chart selection
│   ├── llm.py           # model registry and provider calls
│   └── storage.py
├── models.py, schemas.py, db.py
└── core/                # settings, auth helpers
frontend/                # index.html, app.js, styles.css
Datasets/                # sample data
render.yaml              # Render deployment

---

Built by Viktor Berg · LinkedIn · Released under the MIT License