Intelligent Complaint Triage System
Digithon 2025 Submission | IIT Guwahati
SmartSift is a high-performance, hybrid AI system designed to transform how customer support teams handle large volumes of product complaints. By combining a low-latency CPU-based router with a high-intelligence GPU-based analyst, SmartSift automates complaint triage, reduces operational costs, and converts raw feedback into actionable engineering insights.
Modern customer support systems face three critical challenges:
- Volume Overload – Thousands of tickets ranging from trivial administrative queries to critical hardware failures.
- High Latency – Manual triage delays responses and frustrates users.
- Lost Intelligence – Recurring product issues remain hidden in unstructured complaint text.
SmartSift addresses these issues using a tiered AI architecture that is fast, cost-efficient, and reliable by design.
- Model: Sentence Transformers (MiniLM family)
- Role: Instantly analyzes incoming complaints using semantic similarity
- Outcome: Auto-resolves ~85% of simple queries in milliseconds without cloud inference
- Model: Llama 3 (via Groq API)
- Role: Deep analysis of complex, ambiguous, or high-risk complaints
- Outcome: Identifies root causes, affected devices, and failure patterns
- Real-time complaint testing
- Transparent routing decision (CPU vs GPU)
- Aspect-based sentiment analysis with reasoning
- Upload and analyze CSV datasets
- Automatic separation of trivial vs critical issues
- Efficiency metrics and action tagging
- Review AI-flagged complaints (sarcasm, mixed sentiment)
- Human corrections persisted for future retraining
- Prevents silent model failure and model drift
- Executive-level aggregation of complaint history
- Risk Radar highlighting emerging technical failures
- AI-generated recommended action plans
| Component | Technology |
|---|---|
| Frontend | Next.js (React) |
| Styling | Tailwind CSS |
| Animations | Framer Motion |
| Backend | FastAPI (Python) |
| Local AI | Sentence Transformers |
| Cloud AI | Llama 3 via Groq API |
| Data Layer | CSV / Pandas (MVP) |
Bash
pip install -r requirements.txt
uvicorn app.main:app --reload
Backend runs on: http://localhost:8000
Bash
cd frontend
npm install
npm run dev
Frontend runs on: http://localhost:3000
PostgreSQL + Vector DB integration
CRM connectors (Zendesk, Salesforce)
Voice complaint analysis using speech-to-text models
Continuous retraining pipelines
Nakibul Islam Email: nakibul.sci@gmail.com