BetterMeals

Inspiration

It’s 2025. AI can code, drive cars, write essays, even diagnose disease. But the way we eat is still stuck in the past.

We still scroll endlessly for “healthy recipes,” bargain with cooks, argue with flatmates, improvise grocery lists, and pray dinner isn’t dal-chawal… again. Meanwhile, India is drowning in lifestyle diseases: anaemia in half our women, 101 million diabetics, 70% vitamin D deficiency, and a generation losing its heritage foods.

The system is broken. Your health shouldn’t depend on willpower, Google searches, or guilt. So we built BetterMeals because eating well should finally be as easy as tapping a button.

What It Does

BetterMeals is an AI kitchen butler that runs your home’s entire food workflow on autopilot.

It reads your lab results, preferences, doctor’s prescriptions, and household quirks. Then it generates:

A personalised nutrient plan
A daily 3-dish, ≤45-min cook blueprint
An auto-generated grocery cart
A WhatsApp briefing for your cook each morning

One home, many needs? Our multi-resident optimiser designs overlapping meals so an athlete, a diabetic parent, and a child can all eat right from the same kitchen.

For chronic patients, it enforces the exact diet your doctor prescribes consistently, quietly, and without relying on human discipline.

How We Built It

We combined three layers:

Data Ingestion Engine Labs, biomarkers, symptoms, allergies, budgets, cook constraints.
Nutrient Target Generator + Alpha-Evolve Meal Planner A solver that balances nutrients, preferences, cost, cooking time, and household overlap. It outputs a 3-dish daily plan that meets everyone’s macro/micro needs.
Grocery & Execution Layer A grocery optimiser that converts meal plans into a single household cart, maintains a live kitchen inventory, and auto-orders what’s missing from BigBasket/Zepto. It rounds quantities to pack sizes, handles stock-outs with instant re-planning, and updates pantry levels after each cook shift. The Cook Console then sends step-timed instructions and plating notes to the cook via WhatsApp.

Under the hood: Graph-based optimisation, nutrient density scoring, recipe embeddings, pantry-state tracking, and a resolver that re-plans meals if BigBasket marks an item out of stock.

Challenges We Ran Into

Converting messy lab PDFs into structured biomarkers.
Balancing multiple residents’ needs without exploding menu complexity.
Optimising meals that actually cook within 45 minutes, not theoretical cookbook numbers.
Building meal plans that hit 100+ nutrient targets without boring, repetitive menus.
Handling dynamic constraints like OOS items and cook preferences, then re-planning in under 5 minutes.
One of the hardest parts was automating grocery ordering. We had to rely on a browser-driven agent to navigate BigBasket’s full checkout flow. Browser agents are notoriously brittle tiny UI shifts, popup timing, captcha-like elements, or layout changes could break the flow instantly. Handling these boundary conditions became a constant battle. To make the agent robust, we added visual verification at each step. But every additional screenshot inflated token usage, and a single end-to-end order ended up costing nearly a dollar in reasoning tokens. It worked but it taught us just how fragile and expensive real-world browser automation can be when done with LLM-driven agents.

Accomplishments We’re Proud Of

We built a zero-decision user experience: no app, no friction just WhatsApp and automated execution.
Our solver reliably hits nutrient targets for multiple profiles with only 2–3 dishes per meal.
We created a clinically aligned system: the app can import a doctor’s diet plan and execute it relentlessly.
Early testers reported: “This is the first time healthy eating didn’t feel like work.”

What We Learned

We learned that people don’t fail because of knowledge. They fail because of friction.

Everyone knows they should eat more protein, more greens, less sugar. But between meetings, traffic, fatigue, and cook coordination, even the most motivated person loses.

We learned that the real innovation isn’t a diet plan, it’s removing the human from the hardest parts of behavioral change.

BetterMeals became something deeper: Discipline as a Service.

What’s Next for BetterMeals

Full e-commerce automation: BigBasket/Zepto/Swiggy carts + auto-checkout.
WhatsApp voice assistant for cooks to navigate recipes hands-free.
Continuous biomarker adaptation (CGM, HRV, sleep, energy).
A nutritionist dashboard for chronic-care remote monitoring.
Multi-city pilots in homes with cooks.
And eventually: a nationwide food OS that reduces India’s lifestyle disease burden at scale.

Because millions aren’t unhealthy by choice, they’re unhealthy because the system makes good choices hard.

BetterMeals wants to fix that. To build the intelligence layer between what you think… and what you eat.