Write Elaina's full voice & identity document

Go deeper than the Brand Identity page. This is the document the AI reads first. Needs to cover: how Elaina naturally speaks (cadence, tone, sentence length), phrases she uses vs phrases she'd never use, her sense of humour, what she exaggerates vs underplays, the emotional register she operates in, what she finds cringe, what she genuinely cares about vs what she just posts about.

No code needed Session work Foundational

Map Elaina's full story arc — not just bio, but journey

Document the transformation story in detail: the before (anxious, insecure, lost), the turning point (BBB, reading, community), and where she is now. This story is the emotional backbone of her content. The AI needs to know this arc so it can reference it authentically — not just mention "she built a book club" but understand the why behind all of it.

No code needed Session work Foundational

Define BBB's topic universe — the full map of what we talk about

Every book BBB has read. Every Real Talk theme covered. Every recurring topic: anxiety, identity, leadership, professional English, reading habits, community, Korea expat life. What connects them. What Elaina believes about each one. This becomes the AI's topic pool — a living document it pulls from to make content feel specific and earned, not generated.

No code needed Session work Living document

Write "this is Elaina / this is not Elaina" — real examples

The most powerful calibration tool. Pull 10–15 real captions, scripts, and posts that are 100% Elaina. Then write their "generic AI" equivalents side by side. This gives the AI a concrete before/after to calibrate against every time it writes. This is how you stop it from producing content that sounds like anyone else.

No code needed High impact Calibration doc

Do a full content audit — every post, caption, reel script that exists

Pull every Instagram post, caption, reel, carousel, and event post from @booksbythebeachkr. We need the actual text, the format, and the date. This is the raw archive the AI learns from. We figure out the best way to extract this — API, manual export, or a scraping tool — but we don't need to decide that now. The goal is to have it all in one place.

Needs method decision IG API or manual export High impact

Tag and classify the archive — what performed, what flopped, what Elaina is proud of

Once we have the archive, every post needs three labels: (1) engagement performance (high/med/low), (2) content type (reel/carousel/event/etc.), (3) Elaina's personal rating — does she actually love this, or was it just okay? Performance data + taste data together is more powerful than either alone. This tells the AI what to aim for, not just what existed.

No code needed Manual tagging Elaina input required

Pull social media performance data — reach, saves, comments, follows per post

We need numbers, not feelings. Instagram Insights gives reach, impressions, saves, comments, shares, and profile visits per post. LinkedIn and Threads have their own equivalents. We figure out the most frictionless way to get this data into one place on a regular basis — whether that's a manual monthly export, an API connection, or a third-party tool. The goal: the AI knows what type of content grows the account.

Needs method decision IG Insights API or CSV Recurring process

Design the master context document — what the AI reads before every session

All the knowledge from steps 1–7 needs to be structured into a single, well-organised document the AI loads as context. Not a wall of text — a clear, scannable structure covering: who Elaina is, what BBB is, the topic universe, the voice calibration examples, what's performed well, what the content rules are. We figure out the best format and length to make this fast and effective without burning context window.

Light technical Prompt engineering Critical step

Figure out how Elaina adds new context — the living update flow

After every session Elaina runs — a Real Talk, a BYOB night, a new book read — there's new knowledge the AI should have. We need a zero-friction way for Elaina to add this. A simple input form in this admin? A voice note that gets transcribed? A bullet list she drops in? We decide together what's most natural for how she actually works, then build it into the tool.

Needs design decision UX question Elaina input required

Solve persistence — how the AI remembers approvals, edits, and rejections across sessions

Right now, every conversation with an AI starts from zero. For a long-lived production system, we need memory. When Elaina rejects a draft and says "this is too formal," that feedback needs to persist and affect future generations — not just the current session. We figure out the right approach: a structured feedback log, a fine-tuned system prompt, or a database of corrections that gets folded in. This is one of the most technically involved steps and we can take our time figuring it out.

More technical Long-lived system High impact

Run the voice test — generate 5 pieces and have Elaina score them honestly

Once the knowledge base is built, we generate 5 real pieces — 2 reel scripts, 1 carousel, 1 event post, 1 Threads post — and Elaina scores each one out of 10. Anything below 7 means the context document needs work. We iterate until the baseline score is consistently 8+. This is our definition of "ready to move to Plan 2." No point building the rest of the pipeline until the AI can write like Elaina.

No code needed Elaina scores output Gate for Plan 2

Document what the AI got wrong — build a "correction log" from the first test

Whatever the AI got wrong in step 11 — write it down. Not just "this is bad" but "this is bad because..." and "here is what I would have written instead." This correction log becomes the first entry in the system's living memory. It folds back into the context document. Over time, this log is one of the most valuable things in the whole system — it's where the AI learns Elaina's actual editorial taste, not just her stated preferences.

No code needed Elaina input required Seeds the memory system