Trigent has proved its expertise in building complex AI/ML technologies that are stable and well-integrated.
Founded in 2007, a privately-owned, third-party logistics provider (3PL) specializes in truckload, intermodal freight, and LTL, as well as offers TMS for freight management. Leveraging logistics expertise with its diverse portfolio of affiliated carriers, the 3PL arranges its customer’s freight with maximum efficiency. With branches spread across four states, the organization ensures fun innovative environment where customers, suppliers, and employees collaborate to provide flexible logistical solutions.
The 3PL acts as the link between shippers and carriers and has an existing system to track, monitor, and assign loads. However, the operations pertaining to quotations and pricing were manual. The manual approach to pricing and quotations involved agents relying on calls and emails with high turn-around time, which affected the organization’s ability to focus on strategic business opportunities.
The organization desired to re-imagine its core business operations to enable agility, speed, accuracy, and efficiency.
They aimed to leverage technology to determine prices that offered the most value to shippers while maximizing sales and margins. By mining their business data, they aimed to anticipate market trends, identify micro-segments for target marketing and track operational business patterns for timely asset management.
The organization felt the need to be self-disruptive and aspired to develop a single source of truth to predict the pricing aptly. However, data silos posed a significant challenge. The organization needed to extract petabytes of data from several sources, such as transaction data, freight data, and data from internal systems, to name a few, related to the following parameters:
The organization sought a team of seasoned experts to help them with adequate domain knowledge and technical know-how to build, deploy and integrate the new solution into its core systems.
The organization decided to invest in developing a Machine Learning (ML) model for price prediction to stay competitive in the market while ensuring minimal disruption to its existing business teams. Being a forward-thinking organization aspired to develop an ML model that had no precedence in the market. Several variables needed to be considered, such as Business Analysis, DevOps, Application development, Core (ML) technology, Cloud infrastructure, and QA, to name a few.
The organization partnered with Trigent to leverage its technical knowledge and domain expertise. With the gradual evolution of the project, Trigent’s amoeba business framework enabled the organization to scale up as well as ramp down several critical functionalities in the following stages:
Trigent’s team expeditiously participated in a knowledge transfer effort to fully understand the organization’s internal processes and data flows. The intense 2-week discovery further included a series of collaborative discussions on the technical stack and platform architecture with regard to its pricing model.
With the inputs gathered from the discovery phase, Trigent team hit the ground running and developed an ML technology for price prediction. Based on the requirements defined in Discovery, the AI/ML experts assisted the In-House Engineering team in designing data structures and refining algorithm models for accurate price predictions.
Trigent designed an enterprise-grade ML solution that was well-integrated, seamless, and agile. Python was used to build prototypes, thereby providing the organization with an idea of the custom solution.
Leveraging Trigent’s capabilities in disruptive technologies, the team developed an ML solution for price prediction. The ML model was a Python-based solution, and Jupyter Notebook was added as an interface for the convenient testing of the blocks of code. Additionally, techniques such as bagging and boosting were used to reduce prediction variance. Open-source software such as Pytorch and Sklearn were leveraged to build the ML models.
As a part of data preparation, analysis and management, the team added all the historical pricing data from 37 different internal data fields. In addition, APIs were built in the ML model to capture additional data. Freightwaves was used to capture market intelligence and forecasting intelligence information from DAT, EIA, and McLeod, to name a few. The team built a database leveraging MS SQL. The captured data was then ingested by the predictive engine that provided pricing predictions/outputs. Transformation logic was applied to generate predictive pricing options to a customer quote response.
Azure Machine Learning was used to train, score, deploy, and manage the AI/ML model at scale rapidly and with ease. Azure Blob services was leveraged to build data lakes storing petabytes of data. Additionally, Redis was used to cache prevalent data to improve prediction latency.
Azure monitoring was leveraged for model management. The process of monitoring and retraining the ML model was done in an asynchronous manner. Retaining was triggered on a schedule or when new data would become available by referring to the published pipeline REST endpoint from the previous step. MLFlow was used for the end-to-end management of the ML lifecycle.
By providing accurate pricing predictions to the customer quote response, the AI/ML-powered model enabled the organization to: