Jonatas Pedraza

With my team we worked in a project to improve the performance of the Elasticsearch (1.7.2) cluster thats responsible for provide transactional payments data through many endpoints in our API and the main dashboard used by our partners. We did an study to understand the cluster behaviour and then create strategies to achieve the objectives below:

- Cluster Sizing

The current cluster was not properly sized to the daily demand, with that we worked on resizing the datanodes to boost… Mostrar más

With my team we worked in a project to improve the performance of the Elasticsearch (1.7.2) cluster thats responsible for provide transactional payments data through many endpoints in our API and the main dashboard used by our partners. We did an study to understand the cluster behaviour and then create strategies to achieve the objectives below:

- Cluster Sizing

The current cluster was not properly sized to the daily demand, with that we worked on resizing the datanodes to boost the reading model.
We choose the most appropriate instance type from our cloud provider that fit better in the workload of this type of node.

- High Availability

We added new components to the cluster network architecture to increase service availability like DNS, Load Balancing and Availability Zones.

- Role Segregation

In the current configuration we have masternodes and datanodes, just like the datanodes we choose an instance type that fit better for the masternodes workload.
We use masternodes also as backends of our load balancer which at the moment serves us very well and helped to reduce costs.

- Shard Rebalancing

We rebalance the shards in the cluster to improve stability.

- Monitoring

With these adjustments in the cluster we were able to improve monitoring, which helped our operations team to be more accurate in resolving incidents.

With this new configuration, we achieved high availability, reliability and lower response times. At the moment, all the work performed was done on the cluster infrastructure. We still intend to improve our indexes, replication factors, size of shards so that we can achieve even better rates in both reading and writing model.
Mostrar menos

Golang tools to handle postgres logical replication slots.

Custom PXE server to boot and automate the installation of linux images on the rack blades made by Nokia Networks.
The server runs in a CentOS 7 box , installed and provisioned using Vagrant and Jinja2 to generate all the required files templates.

We love soccer matches in Brazil. So, in this hackday project, we built an automatic goal detector for live video transmissions.
In our pipeline, we do some tasks:
• Fetch video chunks from various live video transmissions. (parsing m3u8 files and getting .TS files)
• Extract image and audio features
• Send these features to a classifier.
• The classifier uses a KNN algorithm to guess if there is a goal. If it is, publish this event in a API

Placeholder Recognizer is a service that identifies if a given frame is a known "frame placeholder", by using image similarity. The application works as a REST API.

Developed with Python, MongoDB and OpenCV.

A daemon to consume images from Globo.com Live Thumbs Service to detect black or freeze video streams and report them.
The main goal of the project is to detect and report quickly any issue that decreases the experience on the live video platform.

Developed with NodeJS and Redis.

A tool to retrieve details about the 's browser and then store the information for inspect and troublesshooting by the team.

Developed with Node.js and MongoDB.

Real-time review about the audience of s in Globo.com video platform.
3-step process: Publish - Aggregation - Presentation

Developed with JavaScript, Python, Kafka and ELK.

The Cockpit is a Web Dashboard Tool for Neemu’s clients. With this tool, Neemu's clients can see the performance of Neemu’s Search Engine and Recommendation System with an excellent graphical experience. This experience allows clients to identify in their website in a short time what are the trending products, the most searched products, the best performance rule for the recommendation system, performance of the autocomplete recommendation among other information. With these metrics, the client… Mostrar más

The Cockpit is a Web Dashboard Tool for Neemu’s clients. With this tool, Neemu's clients can see the performance of Neemu’s Search Engine and Recommendation System with an excellent graphical experience. This experience allows clients to identify in their website in a short time what are the trending products, the most searched products, the best performance rule for the recommendation system, performance of the autocomplete recommendation among other information. With these metrics, the client can make the right decisions to improve the performance of its website using Neemu’s System.

We can split Cockpit in two parts: frontend and backend. In the frontend, we used technologies including HTML5, JavaScript, CSS and PHP. In the backend, a WebService was created and implemented with NodeJS technology. The backend is completely scalable with the objective to meet the demands of the increasing number of Neemu’s clients.

Moreover, another tool was created as a complement of the Cockpit project: the SearchManager. This system allows the to add features into the search engine system to improve the results of search system and prioritize a specific product. The technologies used in this tool are the same as the ones in the Dashboard Tool. Mostrar menos

A bunch of microservices built to collect, extract and analyze data.

Developed with Java, Python, NodeJS and MongoDB.

SaveMe App for Nokia S40 devices.

This was a project to integrates our payments transaction platform with CIP, this was a mandatory regulation of Brazilian government for companies that has financial operations of credit cards.