Case Study
AI-Powered NetSuite Chatbot
Next Gen Web Builders has built an on-premise AI assistant, which enables business analysts and developers to create SuiteQL queries using natural language handoff with full integration into a REST API in NetSuite. This chatbot was developed to address compliance and enterprise-level security, improve access to data, and give fewer reasons to create SQLs manually, as well as accelerate decision-making procedures in departments.
Next Gen Web Builders has developed a secure, knowledge-driven, friendly assistant via DeepSeek LLM, Milvus, SentenceTransformers, and Flask. This chatbot allows the generation of SuiteQL queries in real time and natural language summaries, which will all be hosted locally, allowing complete control over data and infrastructure.
Industry
SaaS / ERP
Technologies
AI, LLM, SuiteQL, RAG
Launched
2014
Country
UAE
About The Project
Conversational Access to NetSuite Data
It is a common occurrence that enterprise users are held up by a bottleneck during data retrieval on the NetSuite. This assistant eliminates such friction by turning plain-language prompts into executable SuiteQL programs, which provide quicker insights without writing a single piece of code.
Overview
We implemented the entire pipeline based on RAG that can be completed by user input and return a list of similar past examples in Milvus, which in turn triggers the DeepSeek LLM on the local machine to create a specific SuiteQL query. The outcome is achieved by performing it with the assistance of NetSuite RESTlet API and summarized with a light language model. The assistant integrates into the current workflows and serves both technical and non-technical users, making the complex reports available within seconds.
Our Challenges
Challenges We Solved with the NetSuite Chatbot
Manual Query Cogs
Teams took a lot of effort to write and validate SuiteQL. Instead of this, we substituted it with the historical context-based AI-generated queries on autopilot.
Unrelated Vector Matches
Inconsistencies were found in the noise in semantic search. We adopted threshold scores and retrieved up to 3 examples for precision.
Schema-Guided Query Accuracy
To avoid generating invalid or imaginary queries, we extracted real table and field definitions from the NetSuite schema (stored in Milvus). This schema context is passed to the LLM during SQL generation—ensuring only valid SuiteQL is produced.
Cleaning up LLM Output
Sometimes the model would include additional text or descriptions. We created a regex cleaner, so whatever would come out would be valid executable SQL.
NetSuite API Liaising
In operation with the NetSuite RESTlet APIs, secure handling of OAuth1 was the essential requirement. We have made sure there is a smooth, secure integration of the chatbot and NetSuite.
On-Premise On-Demand
There was a problem of having the model run locally on limited resources. Our metric was optimizing inference on a CPU and making the system extensible to a GPU when needed.
Solutions
The Results
Our chatbot revolutionized access to data within NetSuite. As teams could speak and understand natural language, and translate it to structured, usable SQL in real time, they were able to spend less time on reporting and dependency on developers. The entire infrastructure, including vector search, LLM query generation, etc., all run on-prem. It is fast, reliable, and the data is kept private.
Real-Time SuiteQL Generation
Vector Matching with Contextual Prompts
Lightweight Summarization of API Responses
Seamless Flask-Based Integration
