Buildable on IONext: A Q&A Assistant for FMCG Field Sales
Use case idea for SIs: a Q&A assistant for SKU catalogs, promo rules and SOPs for FMCG field sales, using RAG and open-weight models, data kept in Jakarta.
FMCG field sales representatives often need a quick answer while standing in front of a store: which SKUs are on promotion, what the minimum order is, and what the return rules say. This pattern uses RAG and open-weight models running in Jakarta, so representatives can ask questions in Bahasa Indonesia and the client's internal documents remain in Indonesia.
Status: a use case idea that can be built on IONext, not a project report. Systems integrators and software houses are welcome to use this idea in their own proposals.
What problem does this assistant solve?
SKU catalogs, promotion programs, and return SOPs are usually scattered across PDFs, spreadsheets, and group chats. New representatives take a long time to learn them all, and outdated versions often remain in circulation. A RAG-based assistant retrieves answers from the latest official documents and cites its sources. As a result, representatives do not have to guess, and supervisors do not have to answer the same question repeatedly.
What does the flow look like on IONext?
- The client's trade marketing team uploads official documents to IONext object storage in Jakarta.
- An ingestion service owned by the SI, running on an ORBIT VM, splits the documents into text chunks and creates embeddings with a self-deployed multilingual open-weight model, such as BGE-M3. The embeddings are stored in a vector database on the same VM.
- Representatives ask questions through the app they already use. The SI's service retrieves the relevant document chunks and sends them to the OpenAI-compatible IONext inference API.
- The model composes a short answer and cites the source documents. As of October 2026, the IONext managed endpoint includes Qwen3.8-27B.
- IONA reads production signals, such as a stalled ingestion pipeline, storage that is nearly full, or a resource exposed to a public IP.
Technical details of this pattern are covered in the article on RAG architecture with data kept in Indonesia.
What should the SI decide up front?
| Decision | Suggested choice | Reason |
|---|---|---|
| Answer source | Only official documents approved by the client | Answers about promotions must be defensible. |
| Expired documents | Store the validity date in metadata and remove the document from the index when the promotion ends | Expired promotions do not appear in answers. |
| Answer not found | The assistant states that it does not know and refers the representative to a supervisor | Safer than letting the model fabricate an answer. |
| Conversation logs | Store in Jakarta and mask store names and owner names | Representatives' questions can contain a store owner's personal data. |
| Data location | Storage, embedding, and inference in Jakarta | Promotion documents are confidential client information, and the client does not need to assess cross-border data transfer risk. |
Where does IONext fit?
IONext provides GPUs in Jakarta, an inference API, the ORBIT workspace for VMs and pipelines, and IONA for reading production signals. The sales app, retrieval logic, document curation with the client, and answer quality evaluation remain the SI's responsibility. IONext does not assess the content of the client's promotion documents. Information on certification and data location is available on the Security and Compliance page.
Try building this pattern: sign up in the IONext Console, then explore ORBIT and IONA. For SIs looking to bring this idea to an FMCG client, further discussion is available through the IONext Partner Program.
Sources
- FlagEmbedding (BAAI), the official repository for BGE embedding models including BGE-M3
- Law No. 27 of 2022 on Personal Data Protection
- Pattern Card: Summarizing Store Visit Reports with an LLM
This article provides general information and is not legal advice. The specific obligations of each project should be reviewed with legal counsel.
IONext is sovereign AI infrastructure hosted in Indonesia, operated by PT Data Awan Nusantara in Jakarta. IONext provides GPU as a Service, managed inference through an OpenAI-compatible API, and open-weight model hosting, certified to ISO/IEC 27001:2022 by TÜV NORD.