AI-focused logistics startup Shipsy launches 'Shipsy Brain' to power AI agents

Enterprise logistics software firm Shipsy on September 1 launched Shipsy Brain in beta, a logistics-focused intelligence layer built to help AI agents make and carry out operational decisions, the Gurugram-based company said.

The product sits inside Shipsy’s AgentFleet platform and can power agents for document validation, address intelligence, anomaly detection, routing and settlement management, the company said.

“The logistics systems are just systems of record. So, the human is taking the decision and recording it into the system. The intelligence is sitting outside somewhere,” Shipsy cofounder and chief executive Soham Chokshi told Economic Times Digital in an interview.


Chokshi said Shipsy’s existing agents had earlier relied on frontier models. Meanwhile, Shipsy Brain uses multiple fine-tuned open-source models for different logistics applications, though he declined to name the underlying models.

Also read: Shipsy pushes AI-driven logistics software as enterprises move beyond legacy systems

“AI in logistics must understand how shipments, drivers, documents, carriers, contracts and many other variables interact with each other and then take the right action. Shipsy Brain brings this operational depth to global supply chains. It is built to help enterprises move beyond dashboards and copilots toward AI systems that can reason, recommend and act within clearly defined business controls.” said Chokshi said.

He added that general-purpose models can struggle with terms and documents specific to logistics. “BOL could mean anything. But we know that it means bill of lading. POD could mean anything, but we know it’s proof of delivery,” Chokshi said.

The models are intended to improve accuracy and response time while lowering operating costs, Chokshi said. “Largely, the three points are speed, accuracy, cost,” he added.

In one live use case, a large quick-commerce retailer deployed an agent to handle orders with incomplete or suspicious customer information, according to Chokshi. Earlier, a driver could remain stuck for about 45 minutes while an employee contacted the customer to check whether the order was genuine, he said. According to Chokshi, now the agent calls the customer, verifies the intent, updates the order and tells the driver whether to proceed. He claimed this cut the resolution time to four minutes and helped recover around 30% of revenue earlier lost when delayed orders were cancelled.

Most customers, however, are not ready to hand every decision to an agent, according to Chokshi. “Most enterprises today, they want a very secure human-in-the-loop framework,” he said. Shipsy and the customer jointly set confidence thresholds, while financial decisions and actions such as cancelling an order remain behind hard guardrails, he added.

The company said Shipsy Brain draws on data linked to more than five billion shipments. This includes more than 50 billion operational events, 1.5 billion automated and human decisions, 100 billion GPS pings and over 5,000 logistics workflows.

Also read: Beyond automation: Why logistics firms are betting on agentic AI

In the company’s benchmark, Shipsy Brain scored 82.2% on overall field extraction, compared with 63.4% for Gemini 3.5, 62% for Gemini 3 and 62.4% for Gemini Pro, the company said. Its score on logistics-domain knowledge was 92.4%, compared with 45.9%, 38.6% and 45.9%, respectively.

Chokshi said the models were tested on real documents from field operations, given the same context and evaluated against verified answers using an identical scoring script. He, however, did not disclose the size of the test set.

According to Chokshi, Shipsy has more than 150 enterprise customers and aims to deploy at least four or five agents at half of them by the end of the year.