Thanks to tools from the open-source community and the power of the cloud, small teams can now implement data analytics solutions that compete with and even outperform those of larger companies. Customization enhances these solutions, driving greater value and results for businesses. Maiolnir aims at being an efficient provider for customized and performant solutions.
About
At Maiolnir we build solutions from data engineering to data science based tools. Our interactions with customers are flexible and dynamic - usually it follows these steps:
Inquire
We dedicate up to a week interacting with the client. In this phase we investigate potential solutions.
Proposal
We offer a detailed plan on what is intended to be implemented for the next 1~2 months.
Hammer
Hands-on time. This period we implement the solution and make adaptations along the way.
Repeat
If after the build period the client wants more adaptations/changes, a new cycle begins with more hands-on work.
Portfolio
Here you'll find all projects we worked on so far. You'll find all sorts of companies and sectors, as well as everything related to data processing and performance optimization.
Dr.Llama
Healthcare Startup
Carrefour: SOT
Retail - online/offline
Wintaylor
Healthcare, Drugstores.
Karza Technologies
Online Retail
Cetrix
Agriculture and Food
GFG: Search
Fashion Retail
BlueMetrics
Finance Credit
Artifacts
On each interaction with companies we may create new tools such as open source repositories that helped us solve a given problem. We call those as artifacts and this section is dedicated to listing all we have produced so far:
Dr.Llama: Video explaining in details how Dr.Llama works, from the Llama model up to the implementation of a customized transformer architecture. The goal of the system is to make automatic interpretation of blood reports.
dbt-flow: This tool was created to help us implement the concept of unit-tests for SQL queries on top of dbt. Mocked data is created for the desired tables and a test is performed on a final node to confirm the transformations worked. The tool is open source.
tfcausalimpact: Many companies have the challenge of making some change to their business and not being able to properly measure the impact of such changes. tfcausalimpact (tfci) was created to help companies to run A/B tests in scenarios where the control group A is not available. The package is based on a original one implemented in R, this time now it's implemented in Python. The tool is open source.
djwto: This tool was implemented when we were interacting with Wintaylor/DPSP. We needed a quick and cheap jwt-based auth mechanism for Django but the ones available at the time didn't offer what we needed so we implemented djwto from scratch. The tool is open source.
pySearchML: Complete elasticSearch based system built on top of Kubernetes and Katib. All the infra and code is open source.
pyClickModels: One of the toughest implementations we ever did. This tool can compute a proxy for the concept of "Judgments" which stablishes how good a given item is to a given search query. The code implemented uses a Dynamic Bayesian Network with binary variables optimized via maximum likelihood. The code is open source.
Pong: During our work with Karza Tech we ventured on implementing some Deep Reinforcement Learning tools. As we had to first develop the know-how, we implemented a system that plays Pong using direct reward optimization. The development is open sourced.
Bigquery: From Zero to Half-Hero: This workshop was developed at GFG Dafiti. It's a hands-on self-paced tutorial for learning how to use BigQuery and more sophisticated techniques.
© Copyright Maiolnir