Over the past months, I have been actively building and refining a broad range of LLM-powered tools and applications. These projects span practical engineering workflows, domain-specific automation, signal-processing support tools, and real-time cloud-connected AI endpoints. Below is a summary of the main developments.
🚀 LLM-Powered Development Highlights
1. Local AI Endpoints on Mobile Devices
I built a small, efficient edge AI endpoint on Pixel, capable of receiving data via Cloudflare Tunnel and producing LLM outputs in real time.
This allows:
- local image/audio inference
- lightweight embeddings
- remote command execution
- secure zero-trust gateway integration
This setup demonstrates how smartphones can act as miniature LLM servers for rapid prototyping.
2. Cloudflare AI Search + RAG Indexing
I implemented Cloudflare AI Search to index my own Python projects and documents, enabling:
- semantic search
- document-aware Q&A
- automatic reranking
- similarity caching for speed
I experimented with pausing/resuming indexing, tuning chunking, and customizing retrieval strategies for different repos.
3. Automated Technical Writing Workflows
Built automated templates using LLMs for:
- LaTeX article/report/book generation
- daily Swedish vocabulary and “Dagens svensklektion” compilation
- CNC machining analytics documentation
- ML books with structured conceptual questions, MCQs, and diagrams
These workflows turn raw Markdown into professional documents efficiently.
4. AI-Enhanced Web Applications
Developed several AI-backed web demos, including:
- a passive towed-array sonar simulation with beamforming, LOFAR, and recognition logic
- a crypto-forensics reporting tool
- interactive DSP teaching tools
The frontend is built with React, while the backend connects to LLM APIs for explanation generation and analysis.
5. LLM-Powered Personal Productivity Tools
Created specialized assistants for:
- structured daily logs and summaries
- job-matching architecture design
- ML engineering
These automations significantly reduce writing time and help maintain consistent documentation quality.
6. LLM-Integrated Dev Environment
Enhanced my Neovim setup with:
- custom AI code completion
- FZF-based search + LLM inline refactoring
- automated README improvement tools
- Python/Rust code generation pipelines
This workflow supports rapid prototyping across signal processing, CNC analytics, and ML engineering.
7. Secure Cloud Workflows
Integrated LLM operations into:
- Cloudflare Zero-Trust
- secure SSH over Tunnel
- GitHub → Cloudflare Pages deployment pipelines
- local → remote artifact synchronization
This enables safe running of LLM services and reproducible deployments without exposing internal networks.
🎯 Summary
These recent LLM developments span practical engineering, cloud automation, signal-processing support, language learning, and secure infrastructure. Together, they form a rapidly developing environment that blends LLMs + DSP + cloud + automation, supporting both my personal workflows and applied industry projects.
More deep-dives will follow in upcoming posts.