TheSyn AI

System overview

TheSyn AI is an academic writing platform for nursing, medical and health-science students and researchers. The index page is the workspace: the single screen where a user selects what they want written, supplies the inputs it needs, spends credits, and receives a finished Word document.

1.1The three pieces

Nothing about the writing lives in Bubble. Bubble collects inputs and sends a mode key; the proxy holds every prompt and decides what to do with that key. This separation is why a new mode can be added to Bubble in two minutes but still needs a proxy edit before it will actually generate anything.

1.2Why the architecture is asynchronous

A full article takes several minutes. A full BSc thesis makes seventeen sequential API calls. Bubble's API Connector times out long before either finishes, so the system never waits for a result.

Instead the submit workflow fires the request and returns immediately. The proxy works in the background and, when finished, calls back into a Bubble backend workflow with the finished text. The page watches the chat record's status field and updates itself. The user can close the tab and come back.

Anatomy of the index page

The page divides into three zones: a fixed navy sidebar, a header bar, and the main working column.

2.1Left sidebar

2.2Header

A reusable element, so it appears identically on every page. Navy background, two-pixel gold bottom border, cream text. Contains the wordmark, a Credits link, and Log Out. The Credits link matters on mobile, where the sidebar is hidden and it becomes the only route to buying credits.

2.3Main column

Cream canvas. From top to bottom:

• Hero band — icon, heading "Transform ideas into publication-ready research", gold rule, subtitle, Support button
• Document surface — appears only when a generation exists; carries the reading toolbar and the Download as Word button
• Composer card — the form, always present
• Status elements — loading text, error text, low-credit warning; each shown by condition

2.4Element and group names

Useful when writing conditions or briefing the Bubble AI Agent.

The dropdown system

Three dependent dropdowns narrow forty-plus modes down to one. Each reads from the same WritingMode option set; what differs is the filter applied.

3.1The chain

Step 1  Category        →  Article Generation
Step 2  Article type     →  Quantitative
Step 3  Mode            →  Full Article (Quantitative) — 15 credits
        Cost            →  15 credits
        Conditional fields appear

3.2How each dropdown is configured

3.3Attributes carried by every mode

These live on the WritingMode option set and drive everything downstream.

Categories and subcategories

4.1The five categories

4.2Subcategories

Subcategories exist mainly inside Article Generation, where the mode count is highest. They split the list by study design before the user ever sees mode names.

• Quantitative
• Qualitative
• Mixed methods
• Review articles
• Reviewer response

Thesis categories are split by chapter rather than design, and the two humanisation tools sit in a category with no meaningful subcategory division.

Mode catalogue and credit costs

5.1Single sections and chapters

5.2Proposals

5.3Full articles

5.4Humanised articles

These run the normal writing pipeline, then pass each section through Ryne for academic humanisation before assembly. The higher price reflects the extra API calls.

5.5Data-analysis articles

Twelve variants. The user uploads a CSV or Excel file; the proxy runs real statistical computation in Python (pandas, scipy, statsmodels) through Anthropic's code execution tool, then writes the article around genuine tables.

5.6Full thesis

Seventeen sequential API calls. Produces title page, Quran verse page, certification, dedication, acknowledgement, abstract, live Word table-of-contents field, abbreviations list, all chapters written then revised then humanised, Python data analysis, and full assembly. Currently hidden in Bubble pending a fix to the Discussion chapter length.

5.7Humanisation tools

The humaniser is offered free for a limited period as the acquisition hook for the advertising campaign. Registration is still required, which is the point — it captures the account.

How the credit system works

6.1Where the balance lives

A single number field, credit_balance, on the Bubble User data type. Every part of the system reads or writes that one field.

6.2The full lifecycle

6.3Spending credits

Two separate mechanisms, and they must agree with each other.

6.4Credit packages

Converted at roughly 1,520 IQD to the dollar. The real saving of Standard over Starter is about 19 per cent, and Premium about 21 per cent — not the 25 and 33 per cent that appeared on earlier versions of the credits page. Academics do the arithmetic; overstated savings read as dishonest.

6.5Displaying costs without maintenance

The credits page cost table should be a Repeating Group bound to All WritingMode, showing each option's Display and credits_cost. Built that way it can never go stale: adding a mode in Bubble makes it appear on the pricing page automatically. Hard-coded rows guarantee the page will be wrong within a month.

Conditional input fields

The composer is adaptive. Once a mode is chosen, only the inputs that mode needs appear. Each field sits inside its own wrapper group with a visibility condition keyed to the selected mode's node_key.

The generation flow, step by step

8.1Submit — Button composer-submit is clicked

8.2The proxy call

Step 9 sends a POST to the /generate endpoint carrying the mode key, all populated input fields, any file URLs, and the chat_id. Authentication is a shared secret header.

POST  https://thesyn-proxy.onrender.com/generate
Header:  x-thesyn-secret
Body:    mode, title, method, file_url, chat_id, ...

The secret header was added after unauthorised direct hits on the endpoint were discovered consuming API budget. Without it the request is rejected.

8.3Background work in the proxy

• Resolve the node_key to a prompt in the PROMPTS object
• For reference modes, run a web-search call first to gather verified sources
• For data modes, download the file and run Python statistics through the code execution tool
• Call Claude to write, then for some modes call again to revise
• For humanised modes, pass each section through Ryne
• Build the .docx through generateWordDoc and encode it as base64

8.4The callback

The proxy POSTs to Bubble's receive_generation backend workflow with the response text, word_base64, and status. That workflow updates the Chat record and deducts credits when status is done.

8.5What the page does

The output surface

Once a generation completes, the result renders as a document rather than a chat log: a white page on the cream canvas, constrained to a readable measure, with a gold rule across the top edge.

9.1Reading toolbar

Open items