I built DeskLens Local, a local AI assistant designed to help small IT support teams resolve recurring technical issues faster.
IT analysts frequently handle repetitive incidents involving VPN connectivity, Outlook, Microsoft Teams, and laptop issues. DeskLens helps by finding similar previously resolved tickets and using AI to draft troubleshooting steps, user responses, and escalation recommendations.
It is designed for small IT teams and managed service providers that want practical AI assistance without the cost and complexity of an enterprise AI platform.
Demo
Project demo: View the source code and setup instructions on GitHub
Code
GitHub Repository — DeskLens Local
How I Built It
DeskLens Local uses Qwen2.5 through Ollama to run an open-weight language model locally. Python handles the application logic, Streamlit provides the user interface, and local text-similarity matching helps identify relevant historical support tickets.
The workflow takes a new incident, searches for similar resolved cases, and uses the local model to generate a draft response and troubleshooting recommendations for an analyst to review.
Once the model and dependencies are installed, the application can run without an internet connection. The model and prompts can also be customized for different IT environments and support workflows.
Why Does Open Innovation Matter?
Open innovation makes DeskLens Local practical for smaller teams that need AI but may not be able to justify recurring API costs or send internal support data to an external AI provider.
Running inference locally helps keep sensitive ticket information on the user's own machine. It also removes per-request model API charges and gives the developer control over model selection, prompts, and future customization.
Compared with a closed, hosted AI API, this approach offers greater deployment flexibility, offline capability, and more control over data handling and operating costs.
The project is an early prototype, so its impact on resolution times and customer demand still needs to be validated through real-world testing.
This is my first time building something around AI and running AI model locally - It was a great learning experience.
Thanks,
Aamir
Top comments (1)
기존에 해결된 VPN·Outlook 티켓을 유사도 검색한 뒤 로컬 Qwen2.5로 대응 초안을 만들고 분석가가 검토하는 흐름이 소규모 지원팀에 실용적이네요. 재사용할 티켓에서 개인 식별 정보와 임시 비밀번호를 먼저 가리고, 초안 옆에 참조한 해결 티켓과 적용 환경을 표시하면 오래된 조치가 그대로 전달되는 일을 줄일 수 있겠습니다.