JioTV+ Search Intelligence
Boosted search engagement 21% and slashed null rates by 65% across 11 languages by designing a search evaluation framework that mapped retrieval and intent detection gaps.
The Challenge
JioTV+'s search experience suffered from high null result rates and poor transliteration across India's 11 major languages. Users couldn't find content they knew existed, leading to frustration and drop-off. The search stack needed systematic evaluation to identify where retrieval and intent detection were failing.
The Solution
- Designed a search evaluation framework mapping Vespa DB retrieval and intent detection gaps via Kibana analytics
- Identified systematic patterns in null results across language variants and transliteration failures
- Drove targeted fixes across the search stack based on data-driven gap analysis
- Implemented improvements across 11 language variants simultaneously
Impact
21%
Search Engagement Boost
65%
Null Rate Reduction
11
Languages Supported
Key Decisions
- Evaluation-first approach: Built the measurement framework before optimizing, ensuring every fix was data-driven rather than intuition-based.
- Language-agnostic architecture: Designed transliteration improvements to work across all 11 languages simultaneously rather than one-by-one.
Tech Stack
Vespa DB
Kibana
Python
SQL
REST APIs