Browse categories
Explore
Fiverr Pro
English
$
USD
I will private, on device ai desktop applications
Shashank
Level 2
About this gig
Your clients won't let you send their data to ChatGPT. I get it. I will build you a Windows desktop application that runs a real LLM Phi-3, Llama 3, Mistral, Qwen, or Gemma entirely on the user's machine
No cloud. No API keys. No data leakage. No recurring OpenAI bills. No GDPR headaches. No HIPAA concerns.
I'm a senior Windows developer (13+ years) currently working on hardware debugging tools at a global silicon company. I understand ONNX Runtime, DirectML, GPU/NPU acceleration, and how to ship a 24 GB model inside an MSIX or Inno Setup installer without breaking it.
Perfect for:
- Law firms handling privileged client data
- Healthcare providers (HIPAA-bound)
- Financial advisors and accountants
- Defence and government contractors (ITAR, CMMC)
- HR tools with employee PII
- Any team bound by GDPR, SOC 2, or internal data-residency rules
- Companies in regions with poor internet connectivity
Tech stack: C# / WPF / WinUI, ONNX Runtime GenAI, llama.cpp, Microsoft.ML.OnnxRuntime, DirectML, Semantic Kernel (local mode), LiteDB for vector storage, MSIX / Inno Setup packaging
Hardware requirements I'll help you plan for: I will recommend minimum specs for your end-users based on the model size
Get to know Shashank
Shashank
Windows Desktop Developer C Sharp, C plus plus , Python , WPF, XAML, AI
Level 2
- FromIndia
- Member sinceJan 2018
- Avg. response time1 hour
- Last delivery3 weeks
Languages
English, German, Portuguese, French
Accomplished Software Developer with 14 years of experience by acting as individual contributor mainly towards Windows Desktop Applications and C++ Programming.
I'm also an AI integration specialist with 4 years of experience in bringing cutting-edge AI capabilities to applications and websites just like yours. With a focus on creating seamless, intuitive interactions, I've helped numerous clients enhance their digital platforms with AI, improving user engagement and operational efficiency.
My Portfolio
FAQ
How good are local models compared to GPT-4
Honestly, not as good at everything — but surprisingly close for many tasks. Phi-3-mini and Llama 3 8B handle Q&A;, summarisation, extraction, and drafting very well. For tasks needing broad world knowledge or complex reasoning, cloud models still lead.
How big is the final installer
Between 2 GB and 8 GB depending on the model. I use installers that download the model on first launch if you prefer a smaller initial download
Will this work on a 5-year-old laptop
Yes, with a smaller model (Phi-3-mini, 3.8B parameters) on CPU — slower, maybe 3–6 tokens per second. For real-time-feeling responses, 16 GB RAM and a modern CPU are recommended
Can it use the NPU on newer Copilot+ PCs
Yes. ONNX Runtime with DirectML can target the NPU on Qualcomm Snapdragon X and newer Intel / AMD NPUs for dramatically faster inference with lower power use
What if I want to update the model later?
The Standard and Premium packages include a model-swap mechanism so you (or your users) can drop in a newer or different model without needing a new installer.
Do you handle fine-tuning
Fine-tuning is a separate engagement. For most use cases RAG (retrieving from your own documents) gets you the same practical result without the cost and complexity of fine-tuning. I'll advise honestly on which you need
Can you sign a HIPAA BAA?
I do not sign BAAs as a solo freelancer, but your app itself can be HIPAA-compliant by design — which is exactly what I build. I'll explain the distinction in our first chat.
What about commercial licensing of the models?
I only use models with permissive licences (Phi-3 MIT, Llama 3 with Meta's commercial licence, Mistral Apache 2.0, Qwen). I will flag licensing implications before we finalise which model to use.
