Browse categories
Explore
Fiverr Pro
English
$
USD
I will fine tune custom llms like llama on your private data
Zubair
About this gig
Stop relying on generic AI. Lets build an AI that knows YOUR business.
Standard models don't know your proprietary data, brand tone, or industry jargon. That is where Custom LLM Fine-Tuning comes in.
Hi, Im Muhammad Zubair, a Software & AI Engineer. I am the founder and creator of Khatta-ka-LLMthe world's first AI language model fine-tuned specifically for the Khattak dialect of Pashto.
Teaching an AI a new indigenous dialect requires elite expertise in NLP, tokenization, and parameter-efficient fine-tuning (PEFT). I bring this exact same engineering to your private data.
My Approach:
- Unsloth & LoRA: I train models (Llama 3, Mistral) 2x faster using 70% less VRAM, saving you massive compute costs.
- Data Formatting: I clean your CSV/JSON data into structured instruction/response pairs (Alpaca format).
- Hugging Face: Direct upload of the trained model to your HF hub.
The DevOps Advantage (Premium):
Most AI devs just give you a file. As a Cloud DevOps Engineer, I containerize your AI (Docker) and deploy it securely to AWS with a FastAPI endpoint so your apps can actually use it.
Message me before ordering to discuss your dataset!
Get to know Zubair
Zubair
Software Engineer , DevOps Architect , AI Infrastructure
- FromPakistan
- Member sinceSep 2022
- Last delivery1 year
Languages
Urdu, English
Hi, I’m Muhammad Zubair — a Software Engineer, DevOps Architect, and Founder of SchoolSMS. I also engineered Khatta-ka-LLM, the world's first AI language model for the Khattak dialect.
I bridge software engineering, bleeding-edge AI, and enterprise cloud systems. Whether you need custom software, Agentic AI (LangGraph/Python), or production-ready cloud architectures (AWS, Kubernetes, OpenShift, Terraform), I build for security, high availability, and scale.
Ready to turn your technical vision into reality? Send me a message today!
My Portfolio
Other AI Development Services I Offer
FAQ
What is the difference between Fine-Tuning and RAG?
RAG gives the AI a database to search for facts. Fine-tuning actually changes the AI's "brain" so it learns a specific tone, format, language, or style (like I did with the Khattak dialect). Often, the best enterprise systems use both!
What format does my data need to be in?
Ideally, your data should be in a CSV or JSONL file, structured as prompt/response pairs (Instruction and Output). If your data is raw text, please message me first so we can discuss formatting it.
Is my private business data safe?
Absolutely. Unlike sending your private data to OpenAI's public ChatGPT API, fine-tuning an open-source model (like Llama 3) means YOU own the model and the data. It remains 100% private.
Why do you use Unsloth for training?
Unsloth is a cutting-edge framework that makes fine-tuning LLMs 2x faster and uses 70% less memory without degrading accuracy. This saves you significant money on cloud GPU costs during the training phase.
Can you deploy the model after training it?
Yes! I am a Cloud DevOps Engineer. If you select the Premium package, I will wrap your model in a FastAPI endpoint, containerize it with Docker, and deploy it to AWS, RunPod, or your preferred cloud provider.
