Built & Maintained with MACH-AI

SnapSite

Free AI Website Builder

> Describe it. AI builds it. Edit by texting.

Paste a URL or describe what you need — a portfolio, blog, wedding page, or small business site — and SnapSite AI builds a complete, mobile-ready, SEO-optimized website hosted instantly. Then keep it fresh by texting changes in plain English.

> highlights

• • AI site generation from a URL or plain-English description
• • Natural language edits — just text the change
• • Instant hosting with free subdomains and custom domains
• • Galleries, contact forms, AI blog, maps and directions
• • Free to start, Pro from $10/month

> services used

• • User authentication (signup, login, sessions)
• • Data API (sites, edits, blog posts)
• • File storage + CDN (site assets, galleries)
• • Payments (Stripe subscriptions)

> finish line

Production release: core flows (build, edit, publish) + auth + payments + persisted data + live custom domains serving real users.

LlamaNet

Free OpenAI-Compatible AI API

> Your Mac is idle 18 hours a day. Let it do something useful.

Turn your Mac, GPU rig, or any computer into a free, public AI API. Run Llama, Gemma, Qwen, Mistral, or DeepSeek behind an OpenAI-compatible endpoint with automatic load balancing and free public URLs.

> highlights

• • 60-second setup with one install command per platform
• • OpenAI-compatible /v1 API — drop-in replacement
• • Free public URLs via Cloudflare tunnels, zero config
• • Model manager: search Hugging Face and download with tracking
• • GPU auto-detect — Apple Silicon, NVIDIA CUDA, or CPU

> services used

• • Data API (nodes, models, request routing state)
• • File storage + CDN (installers, docs)
• • Deployment with auto-scaling

> finish line

Production release: install → register node → serve /v1 chat completions over a public URL + node dashboard + load balancing across nodes serving real API traffic.

TCF — The Content Factory

AI Content Creation Dashboard

> Images, videos, and social content from text.

An AI-powered content factory that turns text descriptions into photorealistic images, motion video, infographics, and social media content — with intelligent prompt enhancement and conversational refinement until the result is exactly right.

> highlights

• • AI image generation — photorealistic, cartoons, memes, infographics
• • Video creation with motion, audio, and cinematic effects
• • Intelligent prompt enhancement with automatic fact research
• • Conversational refinement loop for polished results
• • Session history with iteration tracking and cost visibility

> services used

• • User authentication (signup, login, sessions)
• • Data API (sessions, prompts, iteration history)
• • File storage + CDN (generated images and videos)

> finish line

Production release: core flows (generate, refine, download) + auth + persisted session history + media served via CDN to real users.

CPV Matching API

Enterprise Procurement Classification

> CPV matching rebuilt: 60% to 99% full-hierarchy accuracy at 1B tokens/month.

A CPV (Common Procurement Vocabulary) matching API for government tender discovery at scale. Rebuilt from a single-pass embedding classifier into a structured decision pipeline — rule-based pruning, hierarchical narrowing, targeted LLM reasoning, and confidence scoring with abstention. Enterprise system, shared here as a case study.

> highlights

• • Accuracy scored on the full hierarchy path (Division → Group → Class → Category), not top-code only
• • Bounded latency kept flat as volume scaled — cached CPV tree traversal, deterministic preprocessing
• • Token-bounded prompts with zero-shot → few-shot → deterministic fallback
• • Confidence scoring and abstention instead of forced guesses
• • Steady-state production load of ~1B tokens/month with decreasing cost per match

> services used

• • Data API (CPV vocabulary tree, match history)
• • File storage + CDN (CPV datasets, taxonomy assets)
• • Deployment with auto-scaling and monitoring

> finish line

Production release: hierarchical match API (Division → Category) + confidence scoring & abstention + deterministic fallbacks + deployed endpoint serving steady-state tender traffic (~1B tokens/month). Acceptance criteria: full-path accuracy and bounded latency as volume scales (reported 60% → 99% full-hierarchy accuracy — see case study).