People Element: Survey intelligence

Context

Employee surveys contain free-text comments that need analysis alongside structured answers. Themes and sentiment depend on what people actually wrote and the organization's context. A shared analytics product also needs to keep each client's comments, knowledge and generated insights separate.

System and approach

The NLP pipeline uses Hugging Face models for sentiment classification and an LLM for contextual themes and summaries. Per-client knowledge supplies organizational context. Tenant isolation applies to processing and retrieval, while original comment evidence remains available to support the generated insights.

• Process survey comments inside tenant-isolated jobs and data paths.
• Use Hugging Face models for sentiment classification.
• Supply per-client context to the LLM for themes and summaries.
• Keep original-comment evidence available and avoid cross-tenant embeddings.

Delivered scope

The work covers survey NLP, client-specific context, tenant isolation and summaries grounded in employee comments.

Technology

• Python
• FastAPI
• Hugging Face
• OpenAI
• PostgreSQL
• AWS