R&D Agent

About the Client

Our client, an Indonesian chemical company, develops flavour and fragrance bases for food, beverage and personal care brands. Its R&D team works from a library of more than a hundred raw materials (aroma chemicals, specialty ingredients and carriers), each with its own CAS number, odour profile, purity, cost and regulatory limits.

Before the project, that knowledge lived in spreadsheets, supplier PDFs and the heads of a few senior formulators. Building a new base meant cross-checking materials by hand, re-calculating cost per kilogram whenever an exchange rate moved, and searching through SDS and IFRA documents to confirm a usage level was safe.

The R&D Agent

Dihardja Software built a single workspace where the team’s materials, formulations, customers and reference documents live together, with an AI assistant that works directly on top of that data.

• Raw materials library: every material with its CAS number, barcode, category, physical properties, odour notes and cost per kilogram, with imports, exports and FX rates built in.
• Formulations: compositions are built from the library, costed automatically, and moved through draft, review and approval with a full audit trail.
• Knowledge base: SDS sheets, IFRA standards and internal articles are indexed so the assistant can retrieve and cite them.
• AI assistant: formulators ask questions in plain language, such as “what is the cost per kg of Base A?” or “is 15% Guaiacol within IFRA limits?”. The assistant answers from the team’s own records and sources, and flags exceedances before a formulation goes out.

How Dihardja Software Helps

We started with workshops alongside the formulators and QA team to map how a formulation actually moves from idea to approval, and where time was being lost. From there we designed the data model and the interface together, so the assistant could reason over clean, structured records rather than loose files.

The agent is configurable per team: administrators choose models, set retrieval sources and monitor token usage and cost per user. Every action, whether taken by a person or by the agent, is written to the audit trail, which keeps the system ready for customer and regulatory review. The result is a workspace where a new base can be drafted, costed and safety-checked in a single sitting instead of over several days.