Vision statement 
A local food AI that provides full nutritional value information on any food and can 
generate complete menu proposals based on the user's specification. 
The user can: 
* Create an account and log in. 
* Get complete nutritional value information on any food, including macro nutrients, 
minerals, vitamins, amino acids etc. 
* Get the full nutritional value information for a given food combination, i.e. the user can 
enter the quantities of several foods and get a full nutritional value overview. 
* Search for specific nutrient content and get a sortable list of all foods that contain 
this/these nutrient(s). 
* Store food combinations in named and editable lists. 
* Get menu proposals based on nutritional value and other food constraint input, e.g. 
allergies. 
* Freely chat about anything related to nutrition and get competent answers. 
* The AI must have a local web search tool that it will use to anonymously gather any 
info that it does not have in its local database. 
* No user data leaves the server. 
* The whole project must be in a public listed https://git.btshub.lu repo 
named LocalFoodAI_<your IAM> that is updated in real-time. Make sure to prevent 
confidential data from being uploaded. Add your teacher (id evegi144) as a collaborator. 
Anyone should be able to clone the repo and get his own local food AI running with 
ease. 
* The application must be designed to run entirely on your provided Ubuntu 24.04 VM (8 
vCPUs, 30 GB RAM, no dedicated GPU). You must evaluate and select appropriate 
lightweight, quantized local LLMs (via Ollama) and databases that fit securely within 
this hardware limit. 
Roles 
• You are the product owner and tech lead, whilst Antigravity is your development 
team. This means that you interview the customer to extract the vision, and then 
write the Taiga backlog. 
• Your teacher is the customer and Scrum master. Add your teacher 
(gilles.everling@education.lu) as a stakeholder to your Taiga project. 
Workflow 
• You follow the Scrum manual. 
• You agree on the meaning of "Done" with your teacher.  
• The sprint time box is one week. 
• All meetings are documented in the Wiki. 
• In the Daily Scrum meeting you answer and document in the Wiki the three 
questions: 
o What has been accomplished since the last meeting? 
o What will be done before the next meeting? 
o What obstacles are in the way? 
• Thursday: 
o Sprint Review meeting that lasts at most 15 minutes. 
o Sprint Retrospective meeting that lasts at most 15 minutes. 
o Sprint Planning meeting that lasts at most 30 minutes. You estimate the 
capacity of work, produce the Sprint Backlog and Sprint Goal. 
o Daily Scrum meeting that lasts at most 5 minutes.  
• Friday: 
o Daily Scrum meeting that lasts at most 5 minutes. 
• You perform continuous product backlog grooming. 
• During Sprint Planning, you create User Stories in Taiga and break them down 
into smaller technical Tasks. To execute the work, you feed the overarching User 
Story to the agent for context, but you assign it the specific Task for execution. 
• Before the agent writes the code, it generates an Implementation Plan (an 
Antigravity Artifact). You must set your Antigravity Review Policy to "Request 
Review" (rather than "Always Proceed"). You have to read, evaluate, and approve 
the AI's Python and database logic before it is committed, ensuring you actually 
understand the code being generated. 
• You can use the Gemini CLI to quickly test database queries or prompt logic in 
the terminal before having the Antigravity agent implement it into the main 
codebase. 
• All artifacts generated by Antigravity, e.g. task lists, implementation plans and 
browser recordings, must be attached to the corresponding task and/or user 
story in Taiga. 
• In a real-world Agile environment, code commits must be traceable back to the 
business requirements (User Stories). You must instruct your Antigravity 
agent(s): "When you commit this code, use the commit message: 'TG-<Taiga 
task id>: <Taiga task description>' . Do not attempt to push". If the AI reports a 
'sandbox error' or 'permission denied' when trying to push, it is not a 
technical problem - it is a safety feature. Perform the push manually in the 
terminal, then tell the AI to proceed with the next step. 
• Daily Workflow with Antigravity 
You are the Tech Lead; Antigravity is your Development Team. You do not write 
code; you direct the Antigravity agent. 
o Planning (Thursday): 
• Create User Stories with clear story title and acceptance criteria in 
Taiga for your planned work. To do this you can provide Antigravity 
with the project vision statement and let it generate all the scrum 
artifacts for you and it can even feed them directly into Taiga via 
the Taiga API. 
• In Taiga, break the Story into specific tasks for Antigravity and note 
the task ids (e.g. #3). 
o Development (Thursday/Friday): 
• Orchestrate: Feed your Taiga Task to Antigravity’s Agent Manager. 
Here's an example prompt: "Work on Taiga Task #1. Write a Python 
script to initialize the SQLite database. Once complete, stage, 
commit, and push the files with the commit message: 'TG-1: 
Initialize SQLite DB'." 
Pro-Tip: Create a PROJECT_CONTEXT.md file in the root of your 
repository that outlines your tech stack (Ubuntu, Ollama, SearXNG, 
Database, ...) and the strict "No external APIs" rule. Instruct 
Antigravity to "Read PROJECT_CONTEXT.md before starting" so it 
never hallucinates cloud-based solutions. 
• Review: Set Review Policy to "Request Review." Evaluate the AI's 
Python/Database logic before approving. Before approving the 
commit, verify that: 
• The agent included the TG-<ID> prefix in the commit 
message. 
• The code is strictly local (no external API calls). 
• The logic is correct based on your technical research. 
• Attach: Download the AI's Implementation Plan and attach it to the 
Taiga ticket. 
o The Golden Rule of committing: 
Every time you push code, make sure Antigravity has linked it to a Taiga 
Task ID in the commit message so the webhook works. If that's the case, 
you only need to run git push if not: 
git add . 
git commit -m "TG-<ID>: [Your short description]" 
git push 
(Example: git commit -m "TG-1: Setup SQLite DB structure" ) 
o Verification 
• Taiga: Open your User Story (e.g., #1) and check the History/Git 
tab. You should see your commit message appear there 
automatically. 
• Gogs: Check your Webhook "Recent Deliveries." You should see 
green checkmarks (204 No Content). 
Troubleshooting: 
• If you do not tell the AI to use the TG-<ID> format in the 
commit message, your Taiga board will not update. 
• If you see "Referenced element doesn't exist," verify your 
commit message has the correct TG-<ID> format matching 
an existing Story ID in Taiga.