12M+
active customers worldwide
1600+
natural health and wellness products
57+
years in operation
The client, a leading health and wellness brand serving millions of customers with thousands of products, needed a way to query their own data across structured product spreadsheets, supplier contracts, and campaign documents using plain English, without exposing proprietary data to external AI tools. Inferenz built a conversational AI assistant on a versioned FastAPI backend that routes queries to the right data layer automatically, keeps all data inside the client’s own Snowflake and Neo4j environment, maintains conversation history across multi-turn sessions, and deploys continuously via GitHub Actions with zero manual server intervention.
The client needed AI-powered analytics across structured and unstructured data sources without exposing proprietary product, supplier, and sales data to external platforms. Four core bottlenecks were blocking analyst productivity and decision speed.
Product data, supplier contracts, and campaign reports lived in separate systems and formats. Analysts had no single interface to query structured tables alongside unstructured documents, losing hours on manual cross-referencing with no unified way to ask a question across both data types simultaneously.
Business analysts waited days for budget, supply, and campaign numbers. Without an automated query layer, compiling reports from disparate files created persistent bottlenecks that delayed time-sensitive decisions across merchandising, marketing, and analytics teams.
Sending proprietary product, supplier, and sales data to external AI platforms created unacceptable privacy exposure. The team required a solution that keeps all raw data locked inside their own infrastructure, with Role-Based Access Control ensuring users can only query and view data matching their specific organizational permissions.
Unstructured documents held critical connections between products, suppliers, and campaigns that could not be discovered through flat-file search. Without a knowledge graph layer, analysts were limited to keyword lookup rather than relationship-based exploration across the full body of available data.
Inferenz built a privacy-first conversational AI assistant that routes plain-English queries to the right data layer automatically, keeping all data inside the client's own cloud environment throughout.
Built a FastAPI backend with four versioned API layers (v1 to v4), evolving from basic file upload and querying through to full conversation-history-aware multi-turn AI sessions, letting the Next.js frontend adopt new capabilities without breaking existing integrations.
Implemented a dual-pipeline ingestion layer routing Excel files to Snowflake via Snowpark for fast SQL analytics, and PDFs, Docx, and PPTX files to Neo4j as a knowledge graph via LangChain semantic chunking and LLMGraphTransformer, with parallel threading for high-throughput ingestion at scale.
Powered the query engine with CrewAI and LangChain to route each user query to Snowflake, Neo4j, or both, returning a combined chart JSON payload and natural-language report. V4 in-memory session state tracks the last two turns of conversation context for natural follow-ups, while dedicated service endpoints persist chat titles, session IDs, and full conversation history — giving analysts a seamless multi-session experience without reloading files or repeating context.
Secured all SQL execution inside the client's own Snowflake warehouse via Snowpark so raw data never leaves the client's cloud, with Role-Based Access Control ensuring users only query data within their organizational permissions and every file upload audit-logged in Snowflake.
Automated deployment via GitHub Actions CI/CD on a self-hosted runner, archiving the codebase, installing dependencies in a Python virtual environment, and restarting the supervisor-managed FastAPI service on every push, ensuring zero-downtime releases with no manual server intervention.
Lift in analyst productivity
Answers arrive in minutes instead of days, freeing analysts from manual search and copy-paste across scattered files and disconnected systems.
Insight cycle
Budget, supply, and campaign numbers now surfaced in under ten minutes, dramatically accelerating decisions that previously waited days
Conversational sessions
History-aware AI tracks conversation turns in-session, letting analysts ask follow-up questions naturally without repeating context or reloading files
Data exposure
All SQL executes inside the client's own Snowflake environment. Raw data never leaves the client's cloud, eliminating third-party privacy risk entirely.
Whether you’re starting with data modernization or exploring AI copilots, we’re here to help.
Contact Us
