DEV Community: duchu.nft

SafePlate: an Offline Allergy Meal Planner Running Local Gemma 3 (Built for My Roommate)

duchu.nft — Fri, 02 Oct 2026 07:33:13 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

SafePlate — an allergy-aware meal planner for my roommate, who is allergic
to peanuts and shrimp. He told me weeknight cooking is the worst part of his
day: after work, standing in a noisy dorm kitchen trying to invent dinner that
won't send him to the clinic.

SafePlate takes his allergies, household size, and who he's cooking for
(student / gym-goer / senior), then generates a 7-day Vietnamese dinner plan
with ingredients, steps, a per-day nutrition note, and a consolidated shopping
list. The UI is bilingual (Vietnamese / English).

The whole thing runs on his laptop, offline: open-weight Gemma 3 4B
served locally by llama.cpp. No accounts, no cloud, no tracking, no per-token
bill.

I handed it to him on day one. His words (translated from Vietnamese):

"Pretty useful — normally I have to come up with meals myself, total headache."

His feedback shaped v2 the same evening: "menus need more variety — different
kinds of people, still nutritionally balanced." So I added the audience
selector and the nutrition note. That loop — real person, real complaint, shipped
the same night — is the whole point of this challenge.

Demo

UI (Vietnamese mode, his actual allergies checked):

Result generated 100% locally by Gemma 3 4B:

Quick walkthrough:

No deployed link — and that's deliberate: the app is designed to work with the
wifi off. The demo above ran on a laptop in airplane mode.

Code

https://github.com/duchuAz/safeplate

Stack: one Python file serving the UI + a /plan endpoint, one static HTML
page, one prompt builder. The model backend is stock llama-server with a
local GGUF (ggml-org/gemma-3-4b-it-Q4_K_M).

How I Built It

Open-source AI is not a garnish here — it is the product:

Open weights (Gemma 3 4B, Apache-style Gemma license): downloaded once, runs forever. I started on the 1B quant for speed, then swapped to 4B by changing one file path. Try doing that with a closed API.
Open runtime (llama.cpp): CPU inference on an ordinary Windows laptop (~5–10 tok/s for 4B) — no GPU, no vendor lock-in, fully inspectable.
Deterministic guardrails around a small model: a 4B model sometimes "helpfully" suggests shrimp to a shrimp-allergic user, or returns malformed JSON. So generation runs under a JSON grammar, with retries, plus a rule-based allergen filter (including common misspellings) that drops unsafe dishes before they ever reach the screen. Small open models are honest about their limits — you can see exactly where they end and your code begins.

Everything was built Oct 2–4, 2026, in this repo. Nothing predates the window.

Why Does Open Innovation Matter?

Three reasons, all concrete:

Privacy. Allergy and health data never leaves the laptop. There is no server to breach, subpoena, or acqui-hire. For health-adjacent software, "trust us" is not an architecture — no data is.
Offline reality. His dorm wifi dies regularly. A closed API turns this app into a loading spinner; a local model turns dead wifi into a non-event. I verified with airplane mode on.
Cost and control. $0 to run, forever — this matters for a student. And when a better open Vietnamese-capable model drops, it's a file swap, not a migration.

A closed model would have written prettier Vietnamese on day one. It would also
have uploaded my friend's health profile to someone else's computer, billed per
dinner plan, and stopped working the moment the wifi did. That's a bad trade,
and it's exactly the trade open weights let you refuse.

Prize Categories

Best Use of Gemma

How a Fox and a Single Quote Broke the Critter Gallery — SQLi in Intigriti Challenge 0926

duchu.nft — Tue, 29 Sep 2026 01:07:17 +0000

How a Fox and a Single Quote Broke the Critter Gallery — SQLi in Intigriti Challenge 0926

My write-up for Intigriti Challenge 0926: a cute animal gallery hiding a textbook SQL injection behind a base64 ?pic= parameter. Solved 27/09/2026, submission INTIGRITI-TAPV7AA2 accepted.

The challenge

The Critter Gallery shows one animal per page. The animal is picked via a base64-encoded ?pic= query parameter that decodes to the critter's name, e.g. ?pic=Rk9Y → FOX. Beneath each picture sits a short description pulled from a database. Somewhere on the server, a second table — secret_vault — holds the flag.

The official tip: "the gallery speaks different languages" — the same input is processed differently by two layers of the stack.

First look: two code paths, one input

Requesting FOX vs Fox was the first tell. Both returned the fox description, but the ASCII art differed — one fell back to a default drawing. So the image lookup and the description lookup clearly don't share the same comparison logic: the art is matched one way (PHP, exact/case-sensitive), the description another (MySQL, case-insensitive collation with fullwidth folding). Two languages, exactly as the tip promised.

That mismatch told me the description path talks directly to SQL — worth poking.

Finding the injection: the blank page

Classic quote test, base64-wrapped:

' → blank white page (zero-length response)
\ → blank page too
" → normal page, unaffected

A single quote killing the page while a double quote does nothing means the input lands inside a single-quoted SQL string. The backslash breaking it too confirms the quote isn't being escaped — it's raw string concatenation.

Confirming output: dumping all descriptions

Next, a boolean break-out to prove the query result reaches the page:

?pic=JyBPUiAnMSc9JzE=        # base64(' OR '1'='1)

The description box printed all 8 rows instead of one. Injection confirmed, and the output lands in div.desc.

Column count: UNION needs exactly one

UNION-based extraction needs the right column count. Probing ORDER BY / UNION with 1..n columns showed the query selects a single column (SELECT desc ...), and UNION output renders in the same description div.

Fingerprinting via UNION:

?pic=JyBVTklPTiBTRUxFQ1QgQEB2ZXJzaW9uIC0tIC0=   # base64(' UNION SELECT @@version -- -)

→ MySQL 8.0.46, database critter_gallery, tables animals and secret_vault.

The vault

secret_vault has two columns: id, note. One shot to dump it:

zzz' UNION SELECT GROUP_CONCAT(id,'~',note SEPARATOR '|') FROM secret_vault -- -

?pic=enp6JyBVTklPTiBTRUxFQ1QgR1JPVVBfQ09OQ0FUKGlkLCd+Jyxub3RlIFNFUEFSQVRPUiAnfCcpIEZST00gc2VjcmV0X3ZhdWx0IC0tIC0=

The description box returned:

1~INTIGRITI{01a09f56-74a2-700b-a849-ffe6742327b2}

(flag as rendered; submitted value INTIGRITI{01a09f56-74a2-700b-a849-ffe6742327b2}, accepted ✅)

Lessons

Behavioral diffs are the real tip. The FOX/Fox art mismatch revealed two comparison semantics before I sent a single quote. When one input gets two treatments, one of them is usually injectable.
A blank page is an answer. Zero-length responses on ' and \ (but not ") fingerprint a single-quoted context with no escaping — no error message needed.
Dump small, then go wide. ' OR '1'='1 (8 rows in the desc box) proved output control before I spent requests on schema enumeration.
Know your collations. PHP exact-match vs MySQL case-insensitive/fullwidth-folding comparison is a classic desync primitive — it shows up in auth bypasses too, not just galleries.

Thanks to the Intigriti team for a fun challenge. Full probe trail (14 rounds, from recon to flag) is documented in my notes.