Real projects. Real results.

Non-developer team members wanted to evaluate new GitHub projects but couldn't run them. Cloning, installing dependencies, troubleshooting environments — each repo evaluation was a multi-hour developer task that didn't scale across a curious team dropping links overnight. Kosmos is a Slack bot running on a Mac Studio that automates the entire pipeline. Drop a repo URL in any channel — Kosmos clones it, detects whether it's a web service, CLI tool, or TUI/host-install, and routes accordingly. Web and CLI repos boot in Docker and return as a public + LAN URL (or an interactive shell in the thread); TUI repos install directly on the Mac with a launch command. Same URL while live? It returns the existing one — no rebuild. Containers self-expire at 24h or restart on natural-language request like "bring it back for another day". Beyond that: free-form research chat with shell + web access, REPORT.md generation, Paperclip org/agent management, and a Saturday weekly digest of the team's top finds.

Results: Auto-detects web service / CLI / TUI repos — zero manual config • Web and CLI repos return a live public + LAN URL within seconds • Interactive shell access for CLI tools directly inside Slack threads • 24h container auto-expire keeps the Mac clean; natural-language restore brings repos back • Free-form research chat with shell + web tool access • Auto-generated REPORT.md for each evaluated project • Saturday weekly digest summarising the team's top finds

AI video generation usually needs a $2000+ graphics card. Most people and small businesses can't afford that. I'm building a pipeline that runs entirely on cloud APIs. Write what you want, AI generates images, then converts to video. No expensive computer required.

Results: No expensive hardware needed • Professional-looking videos • Works entirely online • Lip-sync for talking characters

Dev team wanted AI help but generic assistants didn't know their coding standards or project structure. I set up custom Claude Code configs - separate assistants for code review, testing, and docs. Each one knows the team's rules and follows them automatically.

Results: Automatic code reviews • Quick test generation • Consistent coding style • Saves hours of repetitive work

Company needed document search for their whole team. Most AI tools are too technical - regular employees couldn't use them. I built a complete platform with a simple interface. Upload files, ask questions, get answers. No training needed - if you can use Google, you can use this.

Results: Handles 500+ questions daily • Anyone can use it in 5 minutes • Works with PDF, Excel, CSV • Keeps files organized by project

Content creators wanted podcasts but recording took forever and needed voice talent they couldn't afford. I built a tool that converts articles to podcasts. Paste your text, pick the voices, get a podcast with multiple hosts having a natural conversation.

Results: Article to podcast in minutes • Natural-sounding AI voices • Multiple host conversations • 10x faster than manual recording

Healthcare clinic was losing patients who called after hours. Front desk was overwhelmed during the day, couldn't answer everything. I built an AI that answers calls like a real person. It checks the calendar, books appointments, and sends confirmations. Works at 3am just as well as 3pm.

Results: Available 24/7 - never miss a call • 70% fewer missed appointments • Handles 100+ calls per day • Frees up staff for in-person patients

UK team needed meeting transcription. Every tool they tried kept messing up with different accents - too many errors to be useful. I built a transcription system tuned for accent handling. Text appears as people speak, accurate enough that they stopped needing to manually fix transcripts.

Results: 98% accuracy with any accent • Text appears as you speak • No manual corrections needed • Works with any meeting tool

Client had thousands of PDFs with charts and diagrams. Text search was useless - you can't search for "that pie chart showing Q3 sales" with normal search. I built a search that actually understands images. Describe what you're looking for and it finds the exact page with that chart or diagram. No more opening 50 files to find one image.

Results: Search 10,000+ pages instantly • 95% accuracy finding images • No need to open each file • 3x faster than manual search

Client had hundreds of documents. Their team spent hours searching through files to answer simple questions. I built a chatbot that reads all their documents and answers questions. The cool part? It double-checks its own answers before showing them - catches mistakes before the user sees them.

Results: 90% accurate answers • Works with 1000+ documents • Answers in under a second • Saves hours of manual searching