Watson

Watson lets you talk to your data, translating complex numbers into valuable insights that anyone can understand. From analysing your datasets to breaking it down into everyday language and visualisations, Watson helps you do it all—no database language or coding knowledge needed!

Problem

Members of Public (MOPs) and even public officers are not able to use and draw insights from government data to support important life decisions. For people with limited training in database languages or coding, making data accessible and understandable called for an everyday solution.

What we built

With AI becoming an increasingly familiar tool for general use, we created Watson—an AI chatbot that lets anyone ask questions and gain data-driven insights. We focused on HDB resale data, as a relevant dataset for the majority of MOPs.

The general idea was to start as a Large Language Learning (LLM) wrapper—where we feed Watson the metadata of a single table such as the column names, datatypes, and definitions. Users can write a question against the dataset in natural language, and we generate a SQL query using the metadata. We then use this SQL query to query the dataset, and finally, return a natural response!

The result

From the average resale flat size to the month-on-month price trend of flats, Watson analyses the HDB dataset to give users answers to their questions. All in language they can understand!

With additional features, users can also visualise data, explore the full dataset, make use of suggested prompts, and rate Watson’s responses.

What users had to say

GovTech:

Would like to sync up. I'm am leading the development of a similar LLM app in AGD called "AI for FI". "AI for FI" aims to empower officers and data analysts to generate their analysis more efficiently and accurately, by providing them with a more intuitive way to access and analyze finance data.

Workforce Singapore:

I'm building something similar (and coincidentally called SHERLOCK). haha

Would love to trade notes, and thinking on how best to use azure openai's code interpreter.

it's already deployed.

Hackathon feedback:

1. Would be cool to upload my own data to analyse

2. Would be interesting to use internally to deload data teams
   

Team members

Our team included Foong (Project Manager, Lead), Austin, Yong Jie (Engineers), and Rachael (UX Writer).