Data Science and AI implementations that are impact-driven
Enterprise Case Study
Alamtri Resources Indonesia (Adaro)
Mining
Supertype use continual deep learning models on historical data & real-time water level sensors to elevate Adaro's mining logistical operations, enabling a machine learning model to assist in the decision making process that results in a cost savings of USD 2 million annually.
The project spans 1.5 years, and requires Supertype to integrate with multiple IoT (Internet of Things) hardware, set up an on-cloud data warehouse infrastructure for continual data streaming and real-time inference, and deliver real-time water level predictions in the form of proactive alerts and streaming-powered dashboards.
The project spans 1.5 years, and requires Supertype to integrate with multiple IoT (Internet of Things) hardware, set up an on-cloud data warehouse infrastructure for continual data streaming and real-time inference, and deliver real-time water level predictions in the form of proactive alerts and streaming-powered dashboards.
Central Bank of Indonesia
Banking
As the central bank of Indonesia, Bank Indonesia (BI) oversees and upholds the stability of the financial conditions in the country. To accomplish this, BI evaluates its policy-making and regulatory decisions against the general public's attitude towards it. One means of achieving that is through social listening and AI-assisted sentiment analysis.
Gleaning insights from various social media sites is both difficult and time-consuming, particularly since there are 100+ pages, sites, and associated channels to pull these data from. Supertype builds a centralized, queryable API service and automated data pipelines that monitors shifts of public opinion towards monetary policies across hundreds of social media channels, using NLP (Natural Language Processing) techniques to automatically score sentiment and summarize citizen's opinions in real-time. In parallel, Supertype develop a token-based, authenticated API service to enable BI's team to query, build upon, and analyze the data stored in the cloud storage for further analysis (e.g temporal, textual, engagement, sentiment analysis etc.) in a way that is secure, language-agnostic, and highly independent. We then integrate the solution around a custom made web application for the end users of Bank Indonesia to be able to query necessary information at ease.
The high-scale analytics infrastructure reudces the overhead work for its analyst team by 99.93%, accounting for a savings of USD 150,000 annually.
Gleaning insights from various social media sites is both difficult and time-consuming, particularly since there are 100+ pages, sites, and associated channels to pull these data from. Supertype builds a centralized, queryable API service and automated data pipelines that monitors shifts of public opinion towards monetary policies across hundreds of social media channels, using NLP (Natural Language Processing) techniques to automatically score sentiment and summarize citizen's opinions in real-time. In parallel, Supertype develop a token-based, authenticated API service to enable BI's team to query, build upon, and analyze the data stored in the cloud storage for further analysis (e.g temporal, textual, engagement, sentiment analysis etc.) in a way that is secure, language-agnostic, and highly independent. We then integrate the solution around a custom made web application for the end users of Bank Indonesia to be able to query necessary information at ease.
The high-scale analytics infrastructure reudces the overhead work for its analyst team by 99.93%, accounting for a savings of USD 150,000 annually.
Saptaindra Sejati
Maintenance
PT. Saptaindra Sejati appointed Supertype for the design and development of a comprehensive predictive maintenance system that predicts equipment failure and maintenance schedules, reduces unplanned downtime, and reduces reactive maintenance.
The end result is a maintenance cost savings of up to USD 1.2 million annually for each type of heavy equipment across its fleet of mining equipment, and a health index that scores each equipment's health status in real time, using deep learning models that integrate tightly with a host of other internal systems: lab-run oil tests, maintenance logs, fleet management reports, and other proprietary data sources.
The end result is a maintenance cost savings of up to USD 1.2 million annually for each type of heavy equipment across its fleet of mining equipment, and a health index that scores each equipment's health status in real time, using deep learning models that integrate tightly with a host of other internal systems: lab-run oil tests, maintenance logs, fleet management reports, and other proprietary data sources.
BPK Indonesia
Government
Supertype builds a multi-agent architecture using large language models to fully automate the ingestion, extraction, and structuring of more than 5,000 audit and financial reports from Badan Pemeriksa Keuangan (BPK) Republik Indonesia.
Based on preliminary estimates, the Agentic automation workflow eliminates ~40,000 man-hours of manual data entry and review annually, delivering operational cost savings of approximately $1,000,000 each year while freeing expert auditors to focus on high-value strategic analysis.
The multi-agent architecture integrates with a unified web dashboard built upon BPK (Badan Pemeriksa Keuangan Republik Indonesia)'s existing data infrastructure, making it possible for auditors to query in natural language ("show all SOEs with SDG 9 findings") and receive instant responses powered by the ingested data.
Based on preliminary estimates, the Agentic automation workflow eliminates ~40,000 man-hours of manual data entry and review annually, delivering operational cost savings of approximately $1,000,000 each year while freeing expert auditors to focus on high-value strategic analysis.
The multi-agent architecture integrates with a unified web dashboard built upon BPK (Badan Pemeriksa Keuangan Republik Indonesia)'s existing data infrastructure, making it possible for auditors to query in natural language ("show all SOEs with SDG 9 findings") and receive instant responses powered by the ingested data.
Bursa Efek (Indonesia Stock Exchange)
Finance
Supertype embarked on building an ambitious project to completely revolutionize analytics creation for the Indonesia Stock Exchange, incorporating automatic agents that generate beautiful visual-heavy PDF and PowerPoint reports with end-to-end automation. IDX put the estimates of analytics creation time savings at 95%, shaving 450 hours of analytics laborfrom their skilled workforce while increasing the accuracy of stock market reporting from 85% to 99.5%.
To date, more than 2,000 data-heavy financial market reports have been generated through this pipeline, which collectively reaches more than half a million readers every month. All of these reports are custom-branded with IDX's letterhead, and visually consistent with internal branding guidelines — even though the chart generation, tables and analytics are fully programmatically generated by our AI-powered reporting engine.
To date, more than 2,000 data-heavy financial market reports have been generated through this pipeline, which collectively reaches more than half a million readers every month. All of these reports are custom-branded with IDX's letterhead, and visually consistent with internal branding guidelines — even though the chart generation, tables and analytics are fully programmatically generated by our AI-powered reporting engine.
Opera
Internet
Supertype build an audience segmentation and lookalike creation model for Opera, one of the world's largest browser company, thus enabling the publisher to generate user segments for their advertisers. By training on a 256-dimensional device level embedding, we were able to achieve an accuray of 99.57% on 10 million raw data in Opera's internal and external benchmarks.
This model augments AdColony's existing user segmentation capabilities, allowing one of the world's largest mobile ad networks to offer even more precise targeting options to its advertisers, thereby enhancing campaign performance and user engagement.
This model augments AdColony's existing user segmentation capabilities, allowing one of the world's largest mobile ad networks to offer even more precise targeting options to its advertisers, thereby enhancing campaign performance and user engagement.
3Kraters
e-Commerce
The e-commerce category leader in Singapore adopts Supertype's Enterprise Analytics model to revamp its data infrastructure, diagnose its e-commerce tracking leakage, and through the help of our engineers, develop a series of innovations that monitor and optimize the prices of over 300+ SKUs (Stock Keeping Units) that take into account marketplace demand factors, seasonality, and competitor pricing. Equivalent work by a team of analysts and engineers would have cost 3kraters USD 180,000 annually.
Creadits
Technology
Supertype develop a deep learning model to help Creadits — a global creative agency — quantify the creative decay process of their digital assets, and mathematically model when ad creatives plateau in performance. The insights these models are able to yield lead to Creadits producing ad creatives that are 40% more performant than industry peers in its benchmarks.
Zoomd
Internet
Supertype develops an API service that integrates neatly into Zoomd's data infrastructure and intelligently score each device based on their likelihood to respond to a campaign offer and reduce advertising waste by 60% while yielding a lift in response rate of up to 23%.
Toyota Astra Motors
Automotive
Toyota Astra Motor (TAM) is the market leader in Indonesia's car industry with more than 295k+ car unit sales in 2021. As the top car seller, TAM aims to improve after-sales business promotion for a more effective and targeted experience, but the challenge lies in managing the significant annual car sales and customer volume.
TAM granted Supertype access to more than 5 years of raw car services and customer data. In order to automate the data transformation process, Supertype's engineers work with Toyota-Astra Motor's data teams to develop a pipeline that extract, transform, and load (ETL) the data to be consumed in the machine learning model experiments; one that also closely integrates with their Azure Blob Storage and Azure Machine Learning Services, processing more than 2,000,000 automobile sales in a custom-built recommendation engine that saves the company upward of USD 240,000 annually in marketing costs by targeting the right customers with the right promotions at the right time.
We further assist TAM's big data team in validating a few business-related assumptions to enhance the outputs of the machine learning model. Supertype's data scientists engage in machine learning experiments to detect potential customer churn, optimizing promotions for TAM's after-sales business. Simultaneously, our team utilizes unsupervised machine learning to tailor product recommendations to individual customers.
TAM granted Supertype access to more than 5 years of raw car services and customer data. In order to automate the data transformation process, Supertype's engineers work with Toyota-Astra Motor's data teams to develop a pipeline that extract, transform, and load (ETL) the data to be consumed in the machine learning model experiments; one that also closely integrates with their Azure Blob Storage and Azure Machine Learning Services, processing more than 2,000,000 automobile sales in a custom-built recommendation engine that saves the company upward of USD 240,000 annually in marketing costs by targeting the right customers with the right promotions at the right time.
We further assist TAM's big data team in validating a few business-related assumptions to enhance the outputs of the machine learning model. Supertype's data scientists engage in machine learning experiments to detect potential customer churn, optimizing promotions for TAM's after-sales business. Simultaneously, our team utilizes unsupervised machine learning to tailor product recommendations to individual customers.
