MiseEnVue

Inspiration

Most independent restaurants fail within their first five years because of tight margins and reckless spending. Operators often jump on viral social media food trends without realizing they can't afford the new equipment or perishable ingredients.

We wanted to build a tool that figures out a restaurant's safe spending budget first, so every trend recommendation actually fits their real bottom line.

What It Does

MiseEnVue is an AI-powered financial intelligence and food trend execution engine built for independent restaurants.

The system reads an operator's P&L statement to calculate:

Real unit economics
Prime cost health
Safe spending budget

It then cross-references external social media trends against walk-in cooler inventory to generate ROI-filtered, multi-channel marketing campaigns that fit what the kitchen can actually afford to execute.

How We Built It

We built the app using Next.js 16 on the frontend and Supabase for database storage and authentication.

Python scripts ingest live Google Trends and YouTube data, while Gemini 2.5 Flash handles qualitative trend summaries and campaign generation.

To keep financial numbers accurate, we separated all math from the AI. The core domain logic runs strictly through deterministic TypeScript code.

For example, our net budget cap relies on calculated Profit Before Marketing (PBM):

$$ \text{PBM} = \text{Net Profit} + \text{Existing Marketing Spend} $$

$$ \text{Total Safe Budget} = \min( \text{PBM} \times \text{Reinvestment Share}, \text{Revenue} \times \text{Sanity Cap} ) $$

We calculate campaign viability using a strict five-factor scorecard:

$$ \text{Score} = 0.25(\text{Trend})

0.15(\text{Local})
0.20(\text{MenuFit})
0.15(\text{OpsFit})
0.25(\text{Profitability}) $$

Challenges We Ran Into

Accounting exports from tools like QuickBooks and Toast use messy, unpredictable names for line items. We built custom keyword mapping to classify food and labor costs accurately without dropping revenue data.

Preventing AI hallucinations was another hurdle. Early tests showed LLMs making up unrealistic ROI numbers, which forced us to strictly lock all financial calculations inside standard code before sending context to the model.

Accomplishments We're Proud Of

We created a defensible financial model that allocates budgets consistently regardless of prior marketing spend.

Our engine links directly to walk-in cooler inventory, matching trends to available stock with over 80% overlap to help curb food waste.

By pairing Gemini 2.5 Flash with search grounding, we generate complete, localized marketing playbooks in under two seconds.

Finally, we delivered a polished, responsive dashboard styled with modern terracotta and pastel tones.

What We Learned

Restaurant owners do not need another complex analytics dashboard. They need direct, low-risk answers tied to their actual inventory.

Keeping math completely separate from database calls and AI output also made overall system testing significantly faster and more reliable.

What's Next for MiseEnVue

We plan to add direct API integrations with:

Toast
Square
Clover

This will eliminate the need for manual CSV exports.

We are also working on direct integrations with the Meta and TikTok marketing APIs to automatically launch ad campaigns within the computed safe budget.

Lastly, we aim to use multimodal vision features in Gemini 2.5 Flash to automatically update inventory levels by scanning photos of walk-in cooler shelves.