Two Sisters Tinctures is a personal skincare inventory and routine-ordering app built by Earl Grey for her sister. Its central design choice is to let Gemma 4 interpret messy product text, while fixed application rules determine routine order and conflict notes—and a person reviews each proposed product card before it is saved. It is a shelf-management aid, not a medical or allergy checker.
What Two Sisters Tinctures does
In a first-person post published on DEV Community on October 3, 2026, project author Earl Grey describes an app for keeping track of skincare products and how they fit into a routine. A user can type a product name or paste its ingredient list; the app drafts a product card for review. The virtual shelf can show when a product has been open for a long time and display morning and evening routines with notes about products the app says should not be layered. These are the author’s descriptions of the project, not independently verified service behavior.
As an Amazon Associate I earn from qualifying purchases.
Grey says the project began with a practical question: “what should I get?” Her sister already knew her products, and the tracker was meant to record that knowledge and help answer questions about product order. The author reports that her sister responded, “Omg yesssss! 🧪” when told about the project, and later said, “Loving this! A lot of people have questions on which order to use the products too 🤓”. That is feedback from one family member, not broad user validation.
Recommended Free Tools
How the model and rules divide the work
The post identifies the model as Gemma 4 26B A4B (gemma-4-26b-a4b-it), called through Google’s Gemini API. The model reads a product label and drafts structured fields for a card. A validator checks that the response has the expected shape and product type, and limits listed ingredients to those supplied by the user. The user then reviews the proposed card before saving it.
#1 Best Overall
Routine decisions are handled separately. Grey says a plain rules file—not model improvisation—controls product-type rules, morning or evening order, conflicts, and 86 ingredient notes and aliases. In her words: “The model reads. The rules decide.” That is the design described for this project; it is not evidence that its classifications or routine notes are clinically validated.
Why this separation matters
Product-label interpretation and routine guidance are different tasks. The model can turn uneven input into a draft, while explicit rules make the app’s ordering logic more predictable and inspectable. The human review step provides another check before information enters the shelf. None of those steps, by themselves, establishes that an ingredient list is complete or that a routine is suitable for a particular person.
What the author’s small evaluation found
Grey reports testing five products. With both product name and ingredient text, the project classified all five correctly; with ingredient text alone, it classified three of five correctly. In the ingredient-only trials, the BYOMA serum and CeraVe cleanser were each returned as “Moisturizer.” These are results from the author’s five-product set, not an external study or a general benchmark of Gemma 4.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11The result makes the product name a useful part of the input in this project: ingredient text alone did not consistently distinguish product types in those trials. It does not establish expected accuracy on other products, brands, or ingredient lists.
What went wrong during development
The author describes a bug hunt in which a loose substring rule incorrectly pushed a BYOMA serum and a L’Oreal SPF 30 product into night-only use. A comma-splitting issue also damaged an ingredient name. Grey says she changed matching to use whole ingredient names, applied an acid rule of thumb based on position in the ingredient list, and added tests for the affected cases.
Those are reported fixes to this app’s implementation, not general skincare advice. In particular, ingredient-list position and an app’s rule set should not be treated as a substitute for product directions or professional guidance.
What is stored locally, and what is sent out
According to Grey, the saved shelf resides in the browser’s localStorage, with no account or database. But local storage does not mean that nothing leaves the device: she says the product name or pasted ingredient text is sent to the Gemini API so the model can read it. The post says the model key stays on the server and that the service logs reason codes and timings rather than submitted text. These are the author’s descriptions, not an independent privacy or security review, and they do not establish how Google handles API data.
Grey says she checked Google’s paid-service terms in early October 2026 and understood them to say that paid prompts and responses are not used to improve Google products. That is a time-sensitive account of the policy, not an independently verified or permanent guarantee. The post also says the app has daily and per-visitor request limits, including 50 reads a day for the app, 10 per day per visitor, and a per-minute cap; implementation limits may change.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Limits and unfinished features
- Not medical guidance: the author says the app is not a medical tool and does not check allergies. Its scope is a virtual shelf and notes about order and conflicts.
- No photo input yet: the post describes photo input as coming soon and says no photos are sent because the feature is not implemented.
- Saved-card editing is unfinished: the article lists editing product details after saving as not yet available.
- Local model path not tested: Grey describes an Ollama path as a stub, not a tested alternative.
What the project says about testing and cost
The author reports 48 tests, including route tests, and says lint, tests, and npm audit ran in GitHub Actions. These are self-reported development details, not a separate audit of the code or live service.
Grey also reports that the smallest paid Render web service cost $7 per month when she checked in early October 2026. That is a dated author-reported price, not a current quote; hosting costs and service options can change.
Who this project is useful for
For someone building an AI-assisted inventory, the project offers a concrete architectural pattern: use a model to turn imperfect text into a draft, validate the response, apply deterministic rules for consequential app behavior, and require human confirmation before saving. Its reported results also show why relying on an ingredient list alone can be brittle, at least in the author’s five-product evaluation.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →For someone deciding whether to rely on the tracker, the boundaries matter more than the AI label. The app is described as an organizational aid, not an allergy filter, treatment recommendation, or independently validated skincare system. Its shelf is said to stay in the browser, but text submitted for interpretation is sent to an external API.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
