Blockport
blockport
Back to Case Studies
HealthTech

Jubile.ai

Enterprise 500-user pilot · Pre-ChatGPT transformer NLP

Role: Co-Founder & Lead ML Engineer
Period: 2021–2023
500
User enterprise pilot
80%+
Sentiment classification accuracy
Pre-ChatGPT
Transformer NLP in production
Privacy by design
Individuals never scored

The Problem

SMEs struggle to spot early signs of employee burnout. Existing tools depend on annual engagement surveys with response rates below 30% — too slow, too coarse, and triggering survey-fatigue in the people they're meant to help. The signal exists in everyday workplace communication, but tapping it requires both technical sophistication and uncompromising privacy practices.

The Approach

The product measured workforce sentiment and emotional patterns from everyday workplace communication signals — turning subtle data (declining team morale, rising stress in specific functions, communication friction between groups) into actionable insights for HR teams.

Built on **Transformer architectures — the same model family that later powered ChatGPT** — at a time (2021–2022) when very few products were deploying transformer-based NLP in production. ChatGPT wouldn't launch until November 2022; using these models in a live HR product was technical pioneer work, not commodity.

The harder engineering problem wasn't model accuracy. It was building a system where privacy was structural, not policy: all communication data consented at source, anonymized at ingestion, and aggregated only at team or organizational level. Individual employees never scored. No retrofitted GDPR — designed in from day one.

The Outcome

Closed an enterprise pilot with a 500-user company, deploying the system into their day-to-day operations. The pilot validated the core thesis: transformer-based NLP could surface meaningful organizational wellbeing trends without compromising individual privacy.

The product paused as the team's focus shifted in 2023, but the technical foundations — production transformer NLP, privacy-first architecture, HR-facing aggregate analytics — have informed every subsequent AI engagement.

**What this work demonstrates:** Shipping production transformer NLP in 2021 required understanding the architecture, not just calling an API. The skills that separate someone who *understands* AI systems from someone who *uses* AI APIs were built doing work like this. Most consultants in today's market are in the second category.

Want similar results?

Let's discuss how I can help your business achieve measurable outcomes.