QTest Chatbot — Assistente IA para Estudantes
shamilop100/Qtest_chatbot
Assistente IA sem alucinações para institutos educacionais
Ver no GitHubChatbot alimentado por Claude 3 Haiku que responde dúvidas de estudantes sobre cursos, taxas, duração, syllabus e placements de institutos. Construído com FastAPI, oferece interface REST API e CLI, com cache baseado em MD5 para otimizar custos e velocidade.
Instalação
npx convez-ai shamilop100/Qtest_chatbot1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Responder perguntas de alunos sobre cursos e taxas em tempo real
- Reduzir carga do suporte educacional com chatbot 24/7
- Rastrear custos de API e estatísticas de uso com granularidade
Veredito

Solução pronta e bem estruturada para instituições educacionais. Demonstra boas práticas (cache inteligente, rastreamento de custos, controle de alucinações) — útil como referência ou para deploy direto em escolas/centros de treinamento.
README
A tradução para o português deste README ainda está na fila.
🤖 QTest Chatbot
AI-powered student assistant for QTest Software Solution LLP, Kozhikode, Kerala — built on Claude 3 Haiku with a FastAPI backend and a CLI interface.
✨ What it does
QTest Chatbot answers prospective and current student queries about courses, fees, duration, syllabus, placements, and more — instantly, accurately, and only from verified institute data. It never hallucinates facts.
- 🎓 Covers Manual Testing, Java & Selenium Automation, and the combined Master course
- 💬 Dual interface — REST API (for web/mobile frontends) and a terminal CLI
- ⚡ MD5-based response cache cuts repeat API calls to zero
- 💰 Per-call cost tracker in USD and INR
- 🔄 Session memory — remembers the last 2 turns of conversation
- 🛡️ Strict system prompt — stays on-topic, never invents data
🗂️ Project structure
qtest-chatbot/
├── app.py # FastAPI server (/chat, /health, /stats)
├── main.py # CLI interface
├── claude_service.py # Anthropic API wrapper + cost tracker
├── institute_data.py # Single source of truth — all course data
├── assistant_prompt.txt # System prompt template with {context} slot
├── hf_local.py # Keyword-based intent detection (en)
├── config.py # API key + model config (reads .env)
├── requirements.txt
└── .env # Your CLAUDE_KEY goes here (never commit this)
🚀 Quick start
1. Clone and install
git clone https://github.com/your-username/qtest-chatbot.git
cd qtest-chatbot
pip install -r requirements.txt
2. Set up your API key
Create a .env file in the project root:
CLAUDE_KEY=sk-ant-...
Get your key from console.anthropic.com
3. Run the API server
uvicorn app:app --reload
Server starts at http://localhost:8000
4. Or run the CLI
python main.py
📡 API reference
POST /chat
Send a student message and get a reply.
Request
{
"message": "What is the fee for Java Selenium course?",
"session_id": "student_001"
}
Response
{
"reply": "The Java & Selenium Automation course is priced at ₹18,000 for a 3-month duration. For enrollment, call 9961 544 424 😊",
"source": "claude"
}
source will be "cache" for repeated questions — no API cost.
GET /health
{ "status": "ok" }
GET /stats
{
"breakdown": { "cache": 12, "claude": 5 },
"total": 17,
"cache_%": "71%",
"claude_%": "29%",
"cost": {
"calls": 5,
"usd": 0.000312,
"inr": 0.0259,
"avg_inr": 0.0052
}
}
🧠 How it works
Student message
│
▼
Empty check
│
▼
MD5 cache key ──── Cache hit? ──── YES ──▶ Return instantly
(stopwords │
stripped, NO
words sorted) │
▼
ask_claude(msg, history)
├── Trim history to last 4 msgs
├── Inject _SYSTEM prompt (built once at startup)
├── Call Anthropic API (3 retries on rate limit)
└── Track token cost (₹ + $)
│
▼
Save to cache + session
│
▼
Return reply + source tag
System prompt is built once at startup by injecting institute_data.py content into assistant_prompt.txt — so every Claude call carries the full institute context without re-reading files.
📚 Courses covered
| Course | Fee | Duration |
|---|---|---|
| Manual Testing | ₹10,000 | 3 months |
| Java & Selenium Automation | ₹18,000 | 3 months |
| Java Selenium + Manual Master | ₹28,000 | 3 months |
All courses include:
- ✅ 100% Placement Assistance
- ✅ Industry-recognised Certificate
- ✅ Free demo class available
- ✅ Online & Offline batches
- ✅ Night & Sunday batches for working professionals
📍 Emerald Mall, Mavoor Road, Kozhikode, Kerala 📞 9961 544 424 | 🌐 qtestsolutions.com
⚙️ Configuration
| Variable | File | Description |
|---|---|---|
CLAUDE_KEY | .env | Anthropic API key |
MODEL_NAME | config.py | Model (claude-3-haiku-20240307) |
max_tokens | claude_service.py | Max reply length (default: 150) |
temperature | claude_service.py | Creativity (default: 0.3) |
| Cache size cap | app.py / main.py | Max cached entries (default: 500) |
| History window | claude_service.py | Msgs kept per session (default: 4) |
🔌 Intent detection (hf_local.py)
A lightweight, zero-dependency keyword classifier detects query intent before hitting Claude. No GPU, no model downloads — just fast string matching.
Detected intents: greeting, pricing, duration, syllabus, placement, certificate, demo, online_offline, location
Currently available as a utility — ready to be wired in for pre-routing, analytics, or fallback handling.
💡 Extending the project
- Add a new course — edit
institute_data.py. The system prompt updates automatically on next server start. - Change the bot's tone — edit
assistant_prompt.txt. - Connect a frontend — point your React/Flutter app at
POST /chatwith asession_idper user. - Wire in intent detection — call
get_hint(message)inapp.pybefore the cache check to log or route by intent. - Persist sessions — replace the in-memory
_sessionsdict with Redis for multi-worker deployments.
📦 Requirements
anthropic
python-dotenv
fastapi
uvicorn
pydantic
Python 3.10+ recommended.
🔒 Security notes
- Never commit
.envto version control — add it to.gitignore - The API has no authentication by default — add an API key header or rate limiting before exposing to the internet
- CORS is set to
allow_origins=["*"]for development — restrict this in production
📄 License
MIT — free to use and adapt.