Tukey

Tukey: Analytics-in-the-Workflow

Comment

Inspiration

Retail data shouldn’t just describe the past—it should shape the future. Small and mid-sized retailers generate massive POS data daily, yet struggle to convert it into timely, actionable decisions. Tukey was inspired by the need to bridge this gap and make advanced analytics accessible, intuitive, and decision-focused.

What it does

Tukey ingests POS sales data, visualizes key trends through interactive dashboards, applies basic predictive analytics for demand and seasonality, and converts insights into natural-language business recommendations using an LLM. It answers critical questions around stocking, profitability, and customer targeting.

How we built it

Tukey follows a modular, cloud-native architecture. A React-based frontend embeds Tableau dashboards using Tableau’s Embedding API. The frontend communicates with a Python backend built on FastAPI, which serves as the secure API layer.

FastAPI handles authentication, exposes sanitized and aggregated POS data as JSON/CSV endpoints, and integrates with Tableau via REST APIs and Web Data Connectors. This enables live or scheduled data refreshes without manual exports.

An LLM service is integrated into the backend to convert analytical outputs into natural-language business insights. The application is containerized and deployed on cloud infrastructure (e.g., AWS), allowing independent scaling of the web app and Tableau Cloud.

Challenges we ran into

Key challenges included handling inconsistent POS data formats, aligning predictive outputs with business-friendly explanations, and ensuring LLM responses remained grounded in actual analytics rather than generic advice.

Accomplishments that we're proud of

We successfully integrated analytics, visualization, and language intelligence into a single pipeline that converts raw retail data into actionable decisions.

What we learned

We learned the importance of interpretability in analytics, the value of storytelling with data, and how LLMs can act as powerful decision translators when paired with structured insights.

What's next for Tukey

Next, we plan to add real-time data ingestion, advanced forecasting models, customer segmentation, and personalized recommendations—moving Tukey closer to becoming a full-fledged AI retail decision assistant.

Built With

  • built-with-frontend:-react
  • css-backend:-python
  • ecs/fargate
  • fastapi-data-visualization:-tableau-(tableau-cloud)
  • html
  • javascript
  • load-balancer)-data-storage:-relational-database-(rds)-devops:-docker
  • tableau-embedding-api
  • tableau-rest-api-apis-&-integration:-restful-apis
  • web-data-connector-(wdc)-ai-/-llm:-large-language-model-(llm)-for-natural-language-business-insights-authentication-&-security:-jwt-/-oauth2-cloud-&-deployment:-aws-(docker
Share this project:

Updates