Hr ops · Production

Thomas uses Databricks RAG and Vector Search to personalize psychometric assessments at scale

The problem

Thomas' paper-based psychometric assessment model could not scale as their customer base grew, and their legacy platform's enormous content library — built to cover every possible personalization iteration — made it extremely difficult to surface the right insights for each individual client.

First attempt

Thomas' previous approach relied on a labor-intensive model of manually training HR directors and hiring managers to interpret assessments, and a legacy content platform with billions of words covering every possible iteration that could not be efficiently personalized or connected to modern workplace applications.

Workflow diagram · grounded in source
1
User profile submission
trigger
“Because a user profile is completed with Thomas every 90 seconds, it was imperative for them to be able to ingest massive amounts of data more effectively and maintain their leadership in the field of people science”
2
Data ingestion and transformation
integration
“The Databricks Data Intelligence Platform and Mosaic AI tools enabled Thomas to improve their data workflows — from ingestion to transformation to analysis — all within a secure environment”
3
RAG Vector Search retrieval
ai_action
“Using RAG techniques to prompt LLMs via Databricks Vector Search gives Thomas the ability to rapidly get insights from their extensive database”
4
Automated tailored insight generation
ai_action
“this GenAI integration enabled their users to propose queries and receive automatically generated, detailed and tailored insights from unstructured data”
5
Personalized assessment output
output
“We could give our clients the ability to find the answers they need, rather than throwing a 40- to 50-page report at them. This made our content incredibly more dynamic and interactive.”
6
Platform integration delivery
integration
“Databricks also gives Thomas the ability to integrate their new, AI-driven processes into popular platforms like Microsoft Teams. Now, their services are available within workflows clients already use.”
Reported outcome

Thomas integrated GenAI into three platforms in three months, went from proof of concept to MVP in weeks, and now delivers dynamic personalized insights through Vector Search rather than lengthy static reports, resulting in increased user satisfaction and deeper engagement.

Reported metrics
GenAI platform integrations completedthree different platforms
time from proof of concept to MVPweeks
User satisfactionincreased user satisfaction
Content interactivity and personalizationsignificantly more interactive, personalized and efficient
Show all 5 reported metrics
GenAI platform integrations completedthree different platforms
time from proof of concept to MVPweeks
user satisfactionincreased user satisfaction
content interactivity and personalizationsignificantly more interactive, personalized and efficient
employee time on personalized feedbacksaving their own employees valuable time
Reported stack
Databricks Data Intelligence PlatformMosaic AIDatabricks Vector Searchretrieval augmented generation (RAG)natural language processing (NLP)LLMsMicrosoft Teams
Source
https://www.databricks.com/customers/thomas
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Thomas integrated GenAI into three platforms in three months, went from proof of concept to MVP in weeks, and now delivers dynamic personalized insights through Vector Search rather than lengthy static reports, result…

What tools did this team use?

Databricks Data Intelligence Platform, Mosaic AI, Databricks Vector Search, retrieval augmented generation (RAG), natural language processing (NLP), LLMs, Microsoft Teams.

What results were reported?

GenAI platform integrations completed: three different platforms; time from proof of concept to MVP: weeks; User satisfaction: increased user satisfaction; Content interactivity and personalization: significantly more interactive, personalized and efficient (source-reported, not independently verified).

What failed first in this deployment?

Thomas' previous approach relied on a labor-intensive model of manually training HR directors and hiring managers to interpret assessments, and a legacy content platform with billions of words covering every possible…

How is this hr ops AI workflow structured?

User profile submission → Data ingestion and transformation → RAG Vector Search retrieval → Automated tailored insight generation → Personalized assessment output → Platform integration delivery.

Related hr ops cases