Support Ticket Intelligence

End-to-end AI prototype that automates support ticket routing, surfaces similar past cases, and drafts agent responses — deployed live on Railway.

Live demo: lavish-connection-production-5385.up.railway.app  |  API docs: support-ticket-intelligence-production-795d.up.railway.app/docs

---

The Business Problem

Technical support teams spend significant time on manual work that machines can handle:

Problem	This project's solution
Agents manually decide which team should own a ticket	ML routing model classifies tickets automatically with >99% accuracy
Agents re-solve problems that were solved before	Semantic search retrieves the 5 most similar historical cases
New agents don't know how to respond	AI suggestion drafts a response grounded in past resolutions
Managers have no visibility into team performance or SLA health	KPI dashboard shows resolution time, SLA breach rate, and escalation rate

---

Architecture

Synthetic Dataset Generator
        │
        ▼
┌───────────────────────┐      ┌──────────────────────────────────┐
│  2 · Routing ML       │      │  3 · Semantic Search             │
│  TF-IDF + LogReg      │      │  SentenceTransformers + FAISS    │
│  >99% accuracy        │      │  cosine similarity retrieval     │
└──────────┬────────────┘      └─────────────┬────────────────────┘
           │                                 │
           └──────────────┬──────────────────┘
                          ▼
              ┌───────────────────────┐
              │  4 · FastAPI          │
              │  /route  /search      │
              │  /suggest  /status    │
              │  /model-performance   │
              └──────────┬────────────┘
                         ▼
              ┌───────────────────────┐
              │  5 · Streamlit        │
              │  KPI · Model Perf     │
              │  Search · Route       │
              │  AI Suggestions       │
              └───────────────────────┘

---

Dashboard Pages

KPI Analytics

Operational metrics with filters for team, severity, and date range:

• Total tickets, open count, avg resolution time, SLA breach rate, escalation rate
• Weekly volume trend with 4-week linear forecast
• Severity distribution (donut chart)
• Avg resolution time per team (horizontal bar)
• Status distribution, top failure modes, channel breakdown
• Full team summary table with per-team SLA and escalation rates

Model Performance

Transparency into the routing model:

• Hold-out accuracy score
• Top 20 TF-IDF keywords that drive routing decisions
• Row-normalised confusion matrix (recall per class)

Search

Semantic similarity search — find historical tickets most similar to a query, ranked by cosine similarity.

Route

Enter a ticket description and get an instant team assignment with confidence scores for every team.

AI Suggestions

Generates a structured agent response using the 3 most similar historical tickets as context, powered by HuggingFace Mistral-7B (free tier).

---

Quick Start (one command)

Windows:

git clone https://github.com/cirobdomingos-cyber/support-ticket-intelligence.git
cd support-ticket-intelligence
setup.bat

Linux / Mac:

git clone https://github.com/cirobdomingos-cyber/support-ticket-intelligence.git
cd support-ticket-intelligence
bash setup.sh

Then start the two services:

# Terminal 1 — API (auto-generates dataset + trains models on first boot)
cd 4-support-ticket-api
uvicorn main:app --reload --host 0.0.0.0 --port 8000

# Terminal 2 — Dashboard
cd 5-support-ticket-dashboard
streamlit run app.py

Open http://localhost:8501.

First boot: The API auto-generates a 50 000-row synthetic dataset, trains the routing model, and builds the FAISS index automatically. No manual setup step required.

---

Manual Setup (step by step)

# 1. Install dependencies
python -m pip install -r requirements.txt

# 2. Generate dataset
python 1-support-ticket-dataset/generator/generate_dataset.py

# 3. Train routing models
python 2-support-ticket-routing-ml/src/train_baselines.py

# 4. Build semantic search index
python 3-support-ticket-semantic-search/src/semantic_search.py

# 5. Start API
cd 4-support-ticket-api && uvicorn main:app --reload --host 0.0.0.0 --port 8000

# 6. Start dashboard
cd 5-support-ticket-dashboard && streamlit run app.py

---

Module Summary

Folder	Purpose	Status
1-support-ticket-dataset	Synthetic dataset generation (50 k rows, 30+ fields)	✅
2-support-ticket-routing-ml	TF-IDF + Logistic Regression routing model	✅
3-support-ticket-semantic-search	SentenceTransformers embeddings + FAISS index	✅
4-support-ticket-api	FastAPI — routing, search, AI suggestion, KPI data, model performance	✅
5-support-ticket-dashboard	Streamlit — KPI, Model Performance, Search, Route, AI Suggestions	✅

---

API Endpoints

Method	Endpoint	Description
GET	/health	Liveness check
GET	/status	Dataset, model, and index status
POST	/route	Predict team for a ticket description
POST	/search	Find top-K similar historical tickets
POST	/suggest	Generate AI response suggestion
GET	/model-performance	Feature importance + confusion matrix
POST	/train	Retrain routing model
POST	/build-index	Rebuild FAISS semantic index
POST	/generate-dataset	Generate new synthetic dataset
POST	/upload-dataset	Upload a custom CSV dataset

---

Configuration

Environment variable	Default	Description
API_URL	(Railway internal URL)	Dashboard → API URL
HUGGINGFACEHUB_API_TOKEN	—	Required to enable AI Suggestions
HUGGINGFACE_REPO_ID	mistralai/Mistral-7B-Instruct-v0.2	LLM model for suggestions

---

Tech Stack

Python 3.11 · FastAPI · Streamlit · pandas · scikit-learn · SentenceTransformers · FAISS · LangChain · HuggingFace · Docker · Railway

---

About

Built by Ciro Beduschi Domingos as a portfolio project demonstrating end-to-end applied ML and data analytics engineering.