Empowering Non-Technical Users in Pharma With Dashboards and Chatbots Working Together in Databricks

The session shows how disciplined data modeling combined with AI/BI Dashboards can turn that complexity into reliable, self-service insights. The core move is building a thin semantic layer over trusted clinical models, then unifying dashboards and conversational analysis across both structured data and protocol documents. For teams navigating this challenge, understanding what AI analytics is and why it only works with governed data is an essential foundation before choosing any tooling approach.

More data rarely means more clarity in pharma. Precision, provenance, and context determine whether a number can be trusted, and the wrong mix of units, definitions, or collection methods leads to incorrect conclusions. The solution is not a single tool, it is a layered approach.

Why pharma's data context defeats traditional dashboards and generic chat

Clinical trials produce diverse data types, from receptor chemistry and milestones to vitals, labs, and imaging, stored across systems, file shares, and documents. Precision is mandatory, for example mixing glucose units across geographies breaks basic aggregates, and whether a measure was systematically collected or spontaneously reported changes how it can be used.

Dashboards communicate patterns well but only answer questions designers anticipated, and sprawling dashboard estates become hard to maintain. Generic chat is flexible but often misses context, assumes users know how to ask, and can hallucinate. In regulated settings where almost right is not acceptable, neither approach suffices alone.

A thin semantic layer on SDTM with Metric Views

The team starts with tidy, governed data, then maps trial data to CDISC SDTM, the submission standard well suited to aggregation and cross study analysis. Databricks Metric Views sit on top to encode joins, measures, dimensions, friendlier names, and descriptions sourced from the SDTM implementation guide. Shared calculations like unique subject counts become consistent across dashboards and conversational queries.

That layer acts as a governance control. Only approved fields are exposed, so users can analyze by sex, country, or treatment arm without accessing subject level identifiers. Because Metric Views are decoupled from any single study, the same modeling work can be pointed from synthetic data to real studies without rework. To speed documentation, the SDTM implementation guide was packaged as a reusable skill to apply standard descriptions and flag deviations across datasets.

One conversation across tables and protocols

Many questions are answerable with dashboards and a Genie space backed by Metric Views. Other high value questions rely on clinical study protocols, which are hundreds of pages of prose and structured snippets aligned to the M11 template. A knowledge assistant pointed at the protocol folder handles document grounded queries, citing sources.

The most powerful pattern combines both. A knowledge assistant can be added directly to an existing Genie space, so a single question can retrieve relevant protocol criteria, translate them into structured queries against vital signs or other SDTM domains, and return a single answer. Intermediate steps are visible for validation. Users work in one place while the system routes sub tasks to the right source, whether documents or tables.

What changes for self service analytics in regulated settings

Trustworthy self service is a layering problem. Start with clean, modeled data, add a thin semantic layer that encodes definitions and relationships, then build dashboards and conversational agents on the same foundation. With governance applied at the semantic layer, non technical users can explore confidently across both tables and text without sacrificing precision or compliance.

The real unlock: structure first, then conversation

The breakthrough is not a chatbot or a dashboard, it is the discipline of putting a reusable semantic layer over well documented models, then blending structured queries with document retrieval in one conversational experience. Teams get speed and flexibility without compromising the rigor that clinical decisions demand.