Case study
AI productivity & workforce analytics platform
Workforce analytics with activity tracking, AI-driven reporting, intelligent scheduling, and operational insight APIs—FastAPI, PostgreSQL, React, and OpenAI-backed summaries.
Backend lead — pipelines, schema, API contracts
- Core store
- PostgreSQL
- API
- FastAPI + RLS patterns
- UI
- React dashboards
Reference architecture
Problem
- —HR leaders needed consistent metrics across hiring, performance, and attrition signals without exporting sensitive spreadsheets ad hoc.
- —Reports had to be explainable: every aggregate traceable to row-level policies and time windows.
Architecture
- —PostgreSQL models org hierarchy, roles, and fact tables for events (reviews, applications, milestones).
- —FastAPI implements authenticated report endpoints with row-level security aligned to org charts.
- —Batch jobs compute rollups and anomaly flags; OpenAI assists with narrative summaries constrained to structured outputs.
- —React dashboards consume paginated aggregates and drill-down APIs rather than shipping raw tables to the client.
Challenges
- —Correctness vs freshness: nightly rollups vs on-demand recomputation for volatile teams.
- —PII minimization: tokenized identifiers in logs; field-level redaction in export paths.
- —Cost control: cache expensive LLM summaries keyed by dataset version and prompt hash.
Technologies
- —FastAPI
- —PostgreSQL
- —React
- —OpenAI APIs
- —AWS
Engineering decisions
- —Treated LLM outputs as presentation layer only—numeric truth always originates from SQL aggregates.
- —Used pydantic models for every report DTO to keep frontend contracts stable across releases.
- —Adopted feature flags for AI-generated sections so operators can disable them per tenant.
Outcome
- —Scalable reporting APIs with auditable metrics and optional AI narratives.
- —Reduced ad-hoc analyst workload by standardizing definitions in the warehouse schema.