Chitale Bandhu, an Indian manufacturer of sweets, savory snacks, and ready-to-eat foods, deployed the Premisys Agentic AI Platform to investigate fryer-oil consumption and forecast future requirements.
Running on an Intel-powered edge system, the solution combines natural-language interaction with analytical models for anomaly detection and forecasting. Sensitive production data and AI inference remain within the facility.
The challenge: understanding where the oil goes
Fryer oil is both a critical production input and a significant operating cost. Consumption changes with product mix, production volume, fryer conditions, and replenishment cycles.
At Chitale Bandhu, multiple fryers can contribute to a single production output, making it difficult to attribute oil consumption directly to one fryer or product. Operations teams also needed to reconcile recorded usage against load-cell measurements—the sensor readings used to track oil weight.
Previously, teams performed this reconciliation manually. Distinguishing normal process variation from unusual consumption required bringing together fryer, inventory, and production records.
Chitale needed a way to:
Compare recorded consumption with measured inventory.
Investigate unusual usage across fryers and production periods.
Estimate oil requirements for upcoming production.
Support replenishment planning while balancing shortage and excess-inventory risks.
The solution: two assistants on one platform
Premisys deployed two purpose-built assistants using its manufacturing AI platform.
Fryer Oil Usage Analysis Assistant
This assistant compares aggregate oil consumption with load-cell measurements and examines usage at the individual fryer level. It surfaces patterns, discrepancies, and potential anomalies for operators to investigate.
A consolidated dashboard brings together oil balances, consumption trends, inventory measurements, anomalies, and alerts. Users can examine fryer-level trends and export reports for further analysis and communication.
Oil Requirement Forecasting Assistant
This assistant estimates future oil requirements using planned production quantities and product mix.
The forecasts give production, procurement, and inventory teams a common basis for reviewing upcoming requirements and planning replenishment.
Source: the two assistants — solution brief (PDF, page 2) (opens in a new tab)
From an exception to an informed response
The deployment goes beyond displaying historical data. Analytical models identify unusual consumption, while natural-language interaction helps users investigate exceptions and understand their operational context.
Operators can review changes across production periods to determine whether anomalies are becoming more frequent or clustering around particular products or shifts.
The platform also supports follow-up workflows where configured. For example, it can recommend an earlier refill and route the recommendation for approval when consumption exceeds an established operating range.
People remain responsible for consequential decisions. The assistants do not independently change production processes or execute purchases unless those actions have been explicitly configured and authorized by the customer.
Why deploy at the edge?
The solution runs on an Intel® Core™ Ultra Series 3 processor-based edge system. Premisys uses a domain-specific small language model for conversational interaction and specialized AI and machine-learning libraries for analysis and forecasting.
Intel® OpenVINO™ and the Optimum Intel toolkit optimize inference on the processor’s CPU and integrated GPU.
For Chitale Bandhu, this architecture supports:
Local processing of sensitive production information.
Low-latency interaction with operational data.
Operation without continuous cloud connectivity.
A compact on-premise deployment.
Premisys works across existing operational and business systems, adding an intelligence layer without replacing core production infrastructure.
What the deployment enables
Chitale Bandhu now has a unified operational view for reviewing oil consumption, inventory, exceptions, and forecasts.
The deployment provides a foundation for identifying abnormal usage and possible measurement discrepancies earlier, investigating issues through natural-language questions, and making better-informed production and purchasing decisions.
It also establishes a pattern that can be evaluated for other high-value ingredients, fluids, utilities, and production inputs. Any expansion would require validation against the data and operating requirements of the next use case.
Source: business value and path to scale — solution brief (PDF, page 4) (opens in a new tab)
Watch the Webinar - Scaling Industrial Agentic AI: From Proof of Concept to Production Value
In this Intel-hosted webinar on October 13, Janhavi Chitale, CIO of Chitale Bandhu, joined P.K. Gupta, CEO of Premisys, to discuss the deployment and how manufacturers can move from AI experimentation to production value.
Janhavi described what matters most to Chitale:
“For us, the real value isn’t the AI itself. It’s giving our team clearer visibility into what’s happening on the factory floor, reducing time spent analysing reports, and freeing them to focus on solving problems. Importantly, our team remains in control of operational decisions.”
She also outlined the opportunity to expand:
“We’re starting with oil consumption, but we’re already thinking about how this approach could extend to other production lines and high-value ingredients.”
Start with one measurable production problem
Manufacturing AI becomes practical when it addresses a defined workflow with available data and clear acceptance criteria.
Explore how Premisys could help your team analyze material consumption, investigate production exceptions, or forecast operational requirements.