Engineers
Months
We developed a one-stop module for a Malaysian client that provides an interface between multiple, disparate data sources for generating business KPIs.
The client is an HR tech giant that provides its customers with Key Performance Indicators (KPIs).
The customers can now pull data from multiple, disjointed data sources to generate meaningful, holistic insights and KPIs. And the automation through schedulers and data push to BI systems increases the efficiency of the KPI generation.
A Malaysian news agency wanted to automate insight-mining from enormous amounts of data collected from various disparate sources, so they could speed up the generation of useful reports on key societal trends. Their reports were often used by the government to understand trends such as drug usage. Currently, data was being captured from social media, newspapers, and physical scanning of documents, before being manually stored in spreadsheets and analyzed for trends.
The current process was labor-intensive, time-consuming, and inefficient. It was also prone to errors and was not very scalable.
Technical challenges in the existing solution
Manual data entry: Data from different sources was manually entered into spreadsheets.
No standardization: Data came in different formats, sizes, and languages, and needed to be harmonized.
Limitations on the number of requests handled: Current AWS Lambda services had limitations on the number of requests processed at any given time.
Lack of customization: No access control through role management.
Reports lacked insights: No visualization of data to report high-level trends.
TechVariable built a platform that automated data collection and segregation, harmonized different file types into a cohesive set of data files, and enabled the generation of reports with data visualization in real time. With an easy-to-use, interactive, and customized user interface solution, we created three automated data pipelines:
We created three automated data pipelines:
Pipeline 1: A custom scraping engine collects data from all online sources, extracts data from hardcopy formats, PDFs, and jpegs using Optical Character Recognition (OCR), and summarizes lengthy articles according to identified keywords.
Pipeline 2: Video transcription using Google APIs to process video and audio files from YouTube or otherwise.
Pipeline 3: A custom-built mechanism for online sources that segregates hashtags, keywords, stories, etc. for further processing.
Data from multiple social media platforms such as Facebook, Twitter, Instagram, YouTube, and Reddit are aggregated for relevant hashtags and keyword inputs.
With this feature, the client can translate or generate transcripts of videos from YouTube or offline sources in different languages. Lengthy articles are summarized. A data warehouse was created and data was translated into four languages, as per the client’s request. A combination of third-party APIs and a few custom-built APIs have been used for data synchronization and processing.
An elastic search-based system, including Fuzzy search, was built on top of the database. This functionality allows the user to add more keywords as they go, thus enhancing their search experience.
The architecture was designed in such a way that the portal can run for multiple clients on multiple servers.
The solution includes an interactive dashboard built on Tableau that provides the client with a single-window view of important parameters in real time. Auto visualization and reporting were done for batches of old data and real-time data.
Auto-tagging has been enabled using Natural Language Processing (NLP) to segregate and analyze data. The data is then assigned positive, negative, or neutral scores.
