Classic web application with rich UI

Main challenge was to modell the entire logistic process and reduce it's costs.

As a core application for the business, it requires high level of resilience and detailed monitoring.

The UI part was built using React, with Python as a backend language. The system uses a microservice architecture pattern, deployed on AWS Fargate with Terraform. Main data is stored in PostgreSQL, while Clickhouse was used for the metrics. Grafana stack is responsible for monitoring and alerting.

The main idea was to find places where people talk around specific topics. Then use these dialogs first to analyze the potential of the company's services, and then to respond in time and offer these services.

Mostly all parsers and posts classification was made using ChatGPT. Parser execution, data preparation and results interpretation were written on Python.

Deployed on AWS Services with different region zones to increase parsing capacity. Results UI and posts alerts were build on Grafana

Web&Mobile app for realtime analytics

Millions of rows on daily basis. Main challenge was to process all data on the fly, not to lose it and aggregate wisely. The idea was to have analytics as close to "live" as possible

Plenty of connectors written on Python, deployed on AWS Lambda & ECS. All data flows to the Clickhouse through the Kafka

To keep data consistent, separate validation services were implemented in order to find data losses and fix them

Monitoring and alerting based on Grafana stack

Raw data based appication

Development was classified as an experiment, so main challenge was to keep final costs as low as possible with no lack of quality

The big amount of data was planned to come from various sources. In order to fit within the budget, we needed a simple and reliable architecture.

All data connectors are written in Python and run separately on AWS Lambda. They place the data in AWS SQS and then into Clickhouse. Here proceeds all data transformations and aggregations. UI was built based on Grafana. Including graphs, reports and alerts

The idea was to imitate dialog with a guest to analyze waiter's behavior. This is a pretty new type of application, so the challenge was to implement new technologies in the specific case

Core part of the solution was built on ChatGPT. We described, in a collaboration with the restaurant team, different type of guests. How they behave, what are theirs typical orders, spendings, extra needs etc. Then we combined this all together with internal guidelines for waiters to make scoring.

Code was written on Python, deployed on AWS Lambda and telegram messenger was used as UI