03 · Professional Work
Nexodus NLP & Data Quality
Building trust in structured data before analyzing unstructured content
Business value
15+ issues were caught weekly before downstream reporting, and content discussions moved from intuition toward evidence.
01 · Business problem
Daily CRM and social data had no validation layer, while audience perception was hidden inside more than 1,000 captions and comments.
02 · My approach
I separated the work into structured-data validation for reporting trust and BERT-based NLP for recurring audience patterns.
03 · What I built
MySQL and Pandas validation workflows, preprocessing, tokenization, BERT analysis, and stakeholder-ready reporting notes.
What changed my thinking
The corrections, tradeoffsand decisions behind it.
- 01
AI was not the first problem
Basic field and formatting errors could damage reporting before any model was used.
- 02
Themes needed translation
Topic and sentiment outputs mattered only after they were explained in business language.
- 03
Trust came before sophistication
Clean inputs and documented assumptions created more value than model tuning alone.