Athena
The AI Layer That Replaced 90% of Manual Data Work
90%
Workload Eliminated
100s
Daily Queries
0
Hallucinations
Bottom Line
A conversational analytics engine I built inside IOL Pulse now answers hundreds of questions daily — eliminating 90% of the manual data extraction workload.
Situation
Business users and external hotel clients needed data insights constantly but had no direct access. Every request went to the analytics team, who ran SQL queries, built Excel files, and emailed them out. It was a bottleneck that scaled badly.
Challenge
Build an AI-powered analytics layer that non-technical users could query in plain language — with guaranteed accuracy, zero hallucination, and role-aware data access so nobody sees data they shouldn't.
Action
I designed and built Athena: a 10-stage NL2SQL and RAG pipeline combining entity resolution, intent extraction, natural language date parsing, parameterised SQL execution, vector search, and runtime computation — all grounded in live PostgreSQL data. Every answer returns the underlying SQL and data used, so results are explainable and auditable. Deployed inside IOL Pulse with role-based data abstraction, Athena also generates executive summaries and visual insights for every page of the platform.
Result
90% of manual data extraction and Excel delivery to business and external stakeholders is now handled by Athena. Hundreds of questions are answered daily across internal teams and external B2B hotel clients. The analytics team's time shifted from repetitive extraction to higher-value work.