Applied AI Research

Research at EyeAI Solutions

The EyeAI Solutions team conducts applied AI research while developing its Kitchen OS platform, deployed as a living laboratory in a real production kitchen. Every system we study runs where the work actually happens — in live service, under real constraints.

“Research questions come from real service, systems ship to a live kitchen, and we report what we measured — never what we hoped.”

This research is conducted by the EyeAI Solutions team while developing the ReTech Kitchen OS, deployed as a living laboratory at Curry Creations.

22 Integrated operational systems in production
120+ Data models spanning orders, inventory, kitchen operations, and accounting
3 Working papers with fully specified evaluation protocols
Daily Forecast-error (MAPE) measurement an auditable accuracy record
Research Areas

Five Directions, One Kitchen

Each research area starts with a problem from real service and ends with a system measured in production.

Demand Forecasting & Zero-Waste Inventory

Time-series models drive procurement and prep decisions from real order history. Statistical forecasts are combined with live orders, and forecast accuracy (MAPE) is measured daily.

Language Models in Food Operations

LLMs process noisy, multilingual vendor receipts to resolve ingredients under safety constraints, with human review gates before inventory posting — plus conversational ordering.

Kitchen Process Optimization

Scheduling one-to-two-person kitchens using prep graphs, bill-of-materials modeling, and queue-aware pickup ETAs — reducing context-switching for small teams.

Future Direction

Privacy-Preserving Computer Vision

Activity recognition where segmentation removes faces and assigns anonymous track IDs before any analysis. A future direction — not yet deployed.

Human-AI Collaboration in Food Service

Staged-trust rollouts, organizational memory engineering, and adversarial multi-agent system review.

Deployed in Production

Eight Systems Running in a Live Kitchen

Not prototypes. These systems run every day in the living laboratory.

Forecasting

Forecast quality as an auditable daily record

Nightly MAPE persistence with rolling trend analysis.

Recipe-aware procurement forecasting

Ingredient demand prediction with verification gating on pricing.

Kitchen Process

Queue-aware pickup ETAs

Live wait estimates based on the actual cooks working.

Same-station batch sequencing

Clustering cook-together items to reduce transitions.

Language Models

Models only where language helps

Templated notifications on deterministic queues with guardrails.

One canonical receipt pipeline, one human gate

Unified receipt intake with mandatory human review.

Human-AI Collaboration

Adversarial multi-agent review of our own systems

Structured resilience auditing with parallel dimension reviews.

One source of truth for voice and screen

A unified detail builder for voice narration and ticket displays.

Live at Curry Creations

“Language models feed inventory, inventory feeds forecasts, forecasts drive the kitchen — and every link reduces the workload of the person cooking.”

How the architecture fits together
Working Papers

What We Are Writing Up

Three working papers, each with a fully specified evaluation protocol.

Language Models in Food Operations

Never Invent an Ingredient: A Safety-Constrained LLM Entity-Resolution Cascade for Noisy Multilingual Procurement Receipts

A safety-constrained entity-resolution cascade that turns noisy, multilingual vendor receipts into inventory postings without ever inventing an ingredient.

Human-AI Collaboration

The Team of One: Engineered Organizational Memory in Solo-Operator-Plus-AI Software Maintenance

How engineered organizational memory lets a solo operator working with AI maintain production software.

Human-AI Collaboration

First, Do No 86: Safety-Engineering an Always-Listening Voice Agent with Irreversible Authority in a One-Person Kitchen

A failure-modes analysis and staged-trust rollout measurement for an always-listening voice agent with irreversible authority.

Public abstracts are available for all three papers; full drafts are shared with team members during revision.

Team

Who Does the Work

S. Patel

Product Owner

Google Scholar profile

Sharmin Akhter

AI Researcher

Google Scholar profile

Vision

Where This Is Going

Today

AI optimization across the operation

AI optimization across stocking, accounting, prep, and surrounding tasks.

Next

Computer vision, privacy first

Computer vision with privacy-first design.

Always

The standing principle

“The system studies the process, never the person.”

Next Step

See the Kitchen OS in Action

The platform behind this research is the same ReTech Kitchen OS we build for restaurants. Explore the product, or go deeper into the research itself.

See the Kitchen OS in action Read the full research site