A custom assistant trained on context, voice, and brand.
A working scaffold for a personal AI assistant that already knows the things a stranger would have to ask. Built for the days when there are too many open verticals and not enough hours.
- The brand: The Education Alchemist, color palette, voice rules, content patterns
- The day job: Senior TPM, API platforms, regulated fintech context
- The practice: REALTOR® at Corcoran, Providence market knowledge, multilingual service
- The long work: Age of Advocacy, LSAT prep, law school trajectory
- The constraints: spine-safe fitness only, hereditary cholesterol context, time and energy budgets
- The voice: no em dashes, no "passionate about," lowercase confidence
| Mode | Purpose |
|---|---|
| /draft | Drafts in voice: emails, posts, captions, listing copy |
| /audit | Reads inbound copy and flags off-brand phrasing |
| /translate | English ↔ Haitian Creole / Spanish / French (real estate context) |
| /recall | Surfaces context from past planning notes |
| /plan | Weekly planner, prioritized against open verticals |
Most AI assistants start every conversation as strangers. This one doesn't. The cost of re-explaining context is the largest hidden tax on working with AI; this is the fix.
Anthropic API (Claude) · Skills (Anthropic skill files) · Vercel for the web app · Supabase for personal context store · No third-party analytics, ever
[ user input ]
│
▼
[ system prompt + brand brief + voice rules ]
│
▼
[ skill files for vertical-specific tasks ]
│
▼
[ Anthropic API ]
│
▼
[ response, post-processed against brand voice linter ]
This is single-user infrastructure. Personal context lives in a private repo and never enters the public one. The public repo contains the architecture, system prompt scaffolding, and skill file templates only.
Pending. README is the spec; first cut of system prompt and skill files in progress.
Junia Janvier · juniajanvier.com