Voice Viz

VoiceViz is a SaaS analytics platform that turns spoken or typed plain-English questions into database queries, then returns live, interactive dashboards and automated reports. No SQL needed.

Comment

Inspiration

Most people who need answers from data aren't SQL experts. A manager who wants "Monthly revenue by region" usually has to wait for an analyst or learn a BI. tool. We wanted to make asking a database a question as easy as asking a colleague, by typing or just speaking. We also knew that letting users generate Queries open the door to SQL injection, so safety had to be part of the design. from day one.

What it does

VoiceViz is a SaaS analytics platform where users ask questions in plain English. by text or voice. It converts each question into a database command, runs it, and returns an interactive dashboard plus an automated report.

How we built it

  • Frontend: React.js with Tailwind CSS for a clean, responsive dashboard UI.
  • Backend: Python with FastAPI, exposing REST APIs that handle the whole pipeline: input, query generation, validation, execution, and reporting.
  • Pipeline: voice/text input → natural language to database command → threat detection → execution → dynamic visualisation → auto-generated report.
  • Security layer: every generated query passes through a malicious SQL. detector before it reaches the database. On our test set it reached $$\text{Accuracy} = \frac{TP + TN}{TP + TN + FP + FN} \approx 0.92$$

Challenges we ran into

  • Translating vague language into precise queries. Questions like "top customers lately" are ambiguous, so we had to handle unclear intent and Map it to the right tables and columns.
  • Balancing security and usability. A strict detector blocks harmless queries (false positives), while a loose one lets attacks through (false negatives). Tuning toward 92% accuracy meant trading off between the two.
  • Making dashboards dynamic. Query results come back in many shapes, so The frontend had to pick suitable chart types automatically.
  • Voice input. Speech recognition errors can change the meaning of a query. So we had to make the pipeline tolerant of noisy input.

Accomplishments that we're proud of

  • A working end-to-end flow from a spoken question to a finished report
  • A SQL threat detector with 92% accuracy

What we learned

  • How to design clean REST APIs with FastAPI
  • That security is a core feature of any system that generates queries
  • How to turn raw query results into useful, readable visualisations
  • That measuring a classifier honestly (precision, recall, false positives) matters more than a single headline number

What's next for VoiceViz

Multi-database support, saved and shareable dashboards, scheduled reports, and a more robust threat detection model.

Built With

Share this project:

Updates