The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
🚀 Built a Private AI Customer Support Chatbot using Phi-3, Python, .NET, and SQL Server.
The chatbot answers customer queries using a company's internal FAQs, documentation, and support knowledge base without relying on expensive external AI APIs.
⚙️ How it works
• Knowledge base converted into vector embeddings using sentence-transformers
• User query processed via .NET Core API
• Python service performs semantic search to find relevant answers
• Context sent to Phi-3 LLM to generate a response
• Instant reply returned to the chat interface
✅ Fully self-hosted AI solution
✅ Works with private company data
✅ Low operational cost
✅ Easy integration with websites or apps
Excited about building AI-powered support systems and private LLM solutions for businesses.
Feel free to connect — happy to discuss ideas! 🤝
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
Most chatbots fail before the first message
Not because the model is bad because nobody built the parts around it. A generic chatbot answers instantly, sounds like a template, forgets what you said two messages ago, and has no idea when to admit it doesn't know something.
Here's what actually separates a chatbot people trust from one they abandon:
- Grounded answers pulled from your actual docs, not guessed (LangChain)
- Real memory it remembers the conversation, not just the last message (Claude)
- A tone that matches your brand not the default assistant voice
- Knows when to escalate routes to a human instead of bluffing (n8n)
- Every exchange logged, so you can see exactly where it's right and where it's not (Python)
Yes, it's a little slower than a bot that just autocompletes. That's the trade a few hundred milliseconds for an answer people actually believe.
If your chatbot answers fast and wrong, that's not a model problem. That's a missing pipeline.
#AIChatbot #ConversationalAI #RAG #LangChain #ClaudeAI #n8n #PythonDeveloper #AIAutomation #AIEngineer #CustomerExperience #FreelanceDeveloper #ContraCreator
Hi Contra 👋 I'm Sarthak, founder of Ascalify.
Most businesses I talk to don't need "more software." They need less wasted time.
The same customer questions answered 50 times a day. Leads that go cold because no one replied at night. Hours spent copying data between sheets and apps.
That's what we fix at Ascalify. We build simple systems that save time and cut costs:
• AI assistants that answer your customers 24/7
• Automations that remove repetitive manual work
• Custom dashboards and CRMs so everything is in one place
My rule: if it doesn't save you time or money, we don't build it.
I'll be sharing what I build here, including small experiments, real results and lessons learned.
If there's one task in your business you wish ran on its own, tell me in the comments. I'm happy to suggest how to automate it. 👇
A visitor was qualified before sales even saw the chat.
Two quick questions in, the bot already knew this wasn't a browser it was a buyer. The chat routed straight to sales with the context attached, no waiting for someone to notice the lead.
→ Qualifying shouldn't wait for a human to have time to ask the first question.
That instant qualification cuts the lag between chat and sales, keeping the buyer engaged and the handoff smooth.