A growing set of AI architecture and automation work, with an emphasis on systems that are private, inspectable, and useful outside a controlled demo.
The common thread is pragmatic engineering: preserve context, reduce repeated work, expose tradeoffs, and make the system easier to operate after it ships.
Enterprise-grade private RAG pipeline using Qdrant, custom automatic document conversion, VLM visual extraction, local LLMs, dense+sparse hybrid search, and multiple reranking stages.
Built around the hard parts of document intelligence: preserving context during ingestion, making visual information searchable, improving source selection before generation, and keeping sensitive knowledge inside a private/local boundary.
Local-first text expansion concept and landing page focused on privacy, speed, and cross-browser productivity.
The project explores a deceptively simple workflow: turn short triggers into polished, reusable text without making the user give up control of their snippets or daily writing habits.
Reusable format for writing up AI projects with constraints, decisions, and measurable results.
The template keeps future case studies from becoming vague feature lists by forcing the important details forward: problem, constraints, architecture, evaluation, outcomes, and evidence still needed.
Guide-driven project for locally managed Cloudflare Tunnel configuration and simpler RDP/SSH setup workflows.
A practical infrastructure utility focused on repeatable setup, local ownership, and reducing the confusion that can come with secure remote-access plumbing.