Saraan Asim
Saraan Asim
Stage
0→1
Product
Data analytics platform
Dates
Jun to Sep 2025
My part
Built the PII masking, audit logging and test tooling
Stack
React, FastAPI, Flask, PostgreSQL, MongoDB, Snowflake, Selenium

PII-safe analytics

A data analytics platform for a client, with personal data masked by its own service, audit logging and a guest-user system.

0→1

The product

A data analytics platform built for a client. People upload, query and visualise data that includes personal information, on interactive dashboards. It has no public site and the client is not named, so this page describes it at the level of its features.

What I built

PII masking and audit logging. The service that detects personal data and masks it, with an audit log of every data operation it performs.

The test data seeder. A seeder that fills Snowflake with realistic test data, with a performance report.

The load-test bot. A Selenium bot that signs in, navigates to the query dashboard, validates each page and captures screenshots along the way.

How it works

The front end is React with TypeScript. It calls a FastAPI backend that handles the API, uploads, data validation and authentication over PostgreSQL and MongoDB, and the platform includes a guest-user system. The masking service is a separate Flask service. Each service runs in its own container, with separate setups for development and production, and the test tooling works against Snowflake and the running app.

3 decisions

Decisions

  1. PII masking as its own service

    Masking personal data runs as a separate service, away from the main API. Personal data is detected and masked in one place, every data operation it performs is audit logged, and the masking rules can change without touching the app.

  2. Realistic test data at scale

    The hardest constraint was realistic test data at scale. I built a seeder that fills Snowflake with test data shaped like the real thing, so the platform could be exercised at volume without depending on real records.

  3. Load tests through the real interface

    The load-test bot drives the app the way a user does. It signs in, opens the query dashboard, checks that each page loads and captures screenshots at each step.

Feature level

From the project record

Architecture

Web appReactGuest usersAPIFastAPI · FlaskPII maskingown serviceAudit loggingservicePostgreSQLMongoDBSnowflakeseeded test dataLoad testsSelenium botWeb appReactGuest usersAPIFastAPI · FlaskPII maskingown serviceAudit loggingservicePostgreSQLMongoDBSnowflakeseeded test dataLoad testsSelenium bot
Fig. 1. A feature-level view. PII masking runs as its own service, separate from the app.

Result

Outcome

My part ran from June to September 2025. The platform has no public site and the client is not named, so there is no public result to link to. It left the platform with personal data masked by its own audited service, and test tooling that can fill Snowflake and load-test the app again whenever it is needed.

Looking back

What I’d change

I would start load testing earlier. Realistic data at scale was the hardest constraint, and the seeder and the load-test bot are how the platform gets exercised at volume. Having them from the first weeks would have surfaced those limits while the design was still easy to change.