PROJECT 01
Complaint Analytics Platform
Turning hotel complaints into structured insights and actionable plans.
- FastAPI
- React
- LLaMA
- Groq
- Docker
01 Context
Guest complaints across several hotels arrived as unstructured text. Turning them into something management could act on meant reading, categorising and writing reports by hand — slow work that delayed responses. I built a multi-hotel platform that classifies complaints automatically, generates action plans, and tracks the resulting work.
02 Challenges
- Multiple hotels, each with its own staff and access needs.
- Free-text complaints with inconsistent wording and quality.
- LLM calls that can fail, rate-limit or time out under batch load.
03 My role
Designed and implemented the platform end to end: backend API, LLM processing pipeline, frontend, authentication and deployment.
04 Decisions
Two models, two jobs
A smaller LLaMA 3.1 8B handles classification, where speed and cost matter more than prose. LLaMA 3.3 70B is reserved for action plans, where reasoning quality justifies the heavier model.
Batching with retries
Complaints are processed in batches with retry logic, so transient API failures don't silently drop records or block the queue.
Role-based access from the start
JWT authentication with role-based access control keeps each hotel's data scoped to the right people.
05 Architecture
- FastAPI backend exposing complaint, analytics and task endpoints.
- LLM pipeline on Groq: classification → aggregation → action-plan generation.
- React frontend with dashboards and a Kanban-style task board.
- Docker Compose and Nginx for deployment.
06 Results
- Automated action-plan generation reduced reporting time from hours to seconds.
NEXT PROJECT · 02
CleverTap Integration