Designing with intelligence

Designing with intelligence

how we made HungerBox smarter every day

how we made HungerBox smarter every day

The Brief

The Brief

Using AI not as a gimmick, but as an invisible layer of intelligence that transformed how India’s largest cafeterias operate.

Using AI not as a gimmick, but as an invisible layer of intelligence that transformed how India’s largest cafeterias operate.

The Challenge

The Challenge

Simplify a high-volume ecosystem (6 lakh+ daily orders) where every stakeholder was drowning in guesswork, delays, and decision fatigue.

Simplify a high-volume ecosystem (6 lakh+ daily orders) where every stakeholder was drowning in guesswork, delays, and decision fatigue.

The Impact

The Impact

Increasing repeat orders by 22%
Boosting compliance by 21%
Reducing support tickets by 35%.
Improved app rating 2.3 to 4.2

Increasing repeat orders by 22%
Boosting compliance by 21%
Reducing support tickets by 35%.
Improved app rating 2.3 to 4.2

From Orders
to Overload

From Orders
to Overload

At HungerBox, scale was our reality: 6 lakh+ orders every day, across hundreds of cafeterias. On paper, we were thriving. In practice, the cracks showed:

  • Users drowning in endless menus, repeating the same orders, never discovering better options.

  • Admins chasing yesterday’s spreadsheets, always reactive, never proactive.

  • Vendors cooking blind - prepping too much of one dish, not enough of another.

  • Ops teams buried in checklists and audits that nobody trusted.


That’s when we turned to AI - not as hype, but as a design weapon to bring clarity, personalization, and velocity to the entire system.

At HungerBox, scale was our reality: 6 lakh+ orders every day, across hundreds of cafeterias. On paper, we were thriving. In practice, the cracks showed:

  • Users drowning in endless menus, repeating the same orders, never discovering better options.

  • Admins chasing yesterday’s spreadsheets, always reactive, never proactive.

  • Vendors cooking blind - prepping too much of one dish, not enough of another.

  • Ops teams buried in checklists and audits that nobody trusted.


That’s when we turned to AI - not as hype, but as a design weapon to bring clarity, personalization, and velocity to the entire system.

Lorem Ipsum Lorem Ipsum

Lorem Ipsum Lorem Ipsum

Lorem Ipsum Lorem Ipsum

Lorem Ipsum Lorem Ipsum

Finding the gaps: Friction we could Fix

Finding the gaps: Friction we could Fix

We didn’t drop AI blindly. We studied the pain. We mapped the mess.

Methodologies Used:


  • Ethnographic shadowing across 6 cafeterias

  • Feedback sentiment mining from 12k+ app reviews

  • Support ticket tagging and clustering

  • Funnel drop-off heatmaps via Mixpanel

  • 1:1 interviews with vendors, admins, and regular users


We found:


  • Users frustrated by irrelevant menus and blind ordering

  • Admins reactive, not proactive - always waiting on insights

  • Ops struggling with low compliance and no diagnostics

  • Vendors flying blind on what’s selling and why


We didn’t need more features. We needed
intelligence woven into every interaction.

We didn’t drop AI blindly. We studied the pain. We mapped the mess.

Methodologies Used:


  • Ethnographic shadowing across 6 cafeterias

  • Feedback sentiment mining from 12k+ app reviews

  • Support ticket tagging and clustering

  • Funnel drop-off heatmaps via Mixpanel

  • 1:1 interviews with vendors, admins, and regular users


We found:


  • Users frustrated by irrelevant menus and blind ordering

  • Admins reactive, not proactive - always waiting on insights

  • Ops struggling with low compliance and no diagnostics

  • Vendors flying blind on what’s selling and why


We didn’t need more features. We needed
intelligence woven into every interaction.

That's when we asked ourselves:

That's when we asked ourselves:

What if HungerBox could think?

What if HungerBox could think?

Smarter
User Journeys

Smarter
User Journeys

AI meal recommendations

Built a custom engine that uses order patterns, cafeteria menu cycles, and time-of-day data to suggest meals, sides, and combos. → Reduced cognitive load and increased add-on sales


A camera that knew your food

Used AI to scan user-uploaded food images, identify dishes, and generate macro profiles. Integrated nudges and goal tracking into the ordering experience.

  • Image classification model (trained in-house)

  • Nutrition mapping via standardized FSSAI tables


Feedback Routing AI

We built a logic engine to classify feedback based on vendor, item, time slot, and sentiment.

  • Reduced routing errors by 92%

  • Closed-loop feedback time dropped from 3 days to 4 hours

AI meal recommendations

Built a custom engine that uses order patterns, cafeteria menu cycles, and time-of-day data to suggest meals, sides, and combos. → Reduced cognitive load and increased add-on sales


A camera that knew your food

Used AI to scan user-uploaded food images, identify dishes, and generate macro profiles. Integrated nudges and goal tracking into the ordering experience.

  • Image classification model (trained in-house)

  • Nutrition mapping via standardized FSSAI tables


Feedback Routing AI

We built a logic engine to classify feedback based on vendor, item, time slot, and sentiment.

  • Reduced routing errors by 92%

  • Closed-loop feedback time dropped from 3 days to 4 hours

Lorem Ipsum Lorem Ipsum

Lorem Ipsum Lorem Ipsum

Lorem Ipsum Lorem Ipsum

Smarter
Operations:
Wings App

Smarter Operations:
Wings App

Checklist AI Validation

Ops managers uploaded kitchen and hygiene checklists. We layered AI to validate completion with image proof and anomaly detection.

  • Flagged recurring misses

  • Scored compliance weekly per location


Response Analysis Dashboard

AI-generated insights helped cluster issues and suggest targeted interventions for each cafeteria.


The Impact:

  • Compliance improved by 21%

  • Reduced field audit costs by 40%

Checklist AI Validation

Ops managers uploaded kitchen and hygiene checklists. We layered AI to validate completion with image proof and anomaly detection.

  • Flagged recurring misses

  • Scored compliance weekly per location


Response Analysis Dashboard

AI-generated insights helped cluster issues and suggest targeted interventions for each cafeteria.


The Impact:

  • Compliance improved by 21%

  • Reduced field audit costs by 40%

Lorem Ipsum Lorem Ipsum

Lorem Ipsum Lorem Ipsum

Smarter
Admins

Smarter Admins

Café Intelligence Console

An AI-powered module where admins ask plain questions and get live data:

  • “What’s my least busy hour this week?”

  • “Which vendor’s NPS dropped?”

  • “What feedback themes are trending?”


Built as a natural-language prompt engine + chart generator

The Impact:

  • Reports generated 8x faster.

  • Admin logins went up by 37%

Café Intelligence Console

An AI-powered module where admins ask plain questions and get live data:

  • “What’s my least busy hour this week?”

  • “Which vendor’s NPS dropped?”

  • “What feedback themes are trending?”


Built as a natural-language prompt engine + chart generator

The Impact:

  • Reports generated 8x faster.

  • Admin logins went up by 37%

Lorem Ipsum Lorem Ipsum

Smarter
Vendors

Smarter Vendors

Sales Forecasting + Menu Intelligence

We provided vendors with demand predictions, dish-level sales trends, and macro suggestions:

  • “Prepare more dal on Monday”

  • “Run an offer on low-sell items at 3 PM”


Vendor AI Assistant (Chat)

Conversational UX layered on top of analytics.


The Impact:

  • Prep errors dropped by 28%

  • Vendor satisfaction score up +17% over 6 months

  • For the first time, kitchens weren’t reacting. They were predicting.

Sales Forecasting + Menu Intelligence

We provided vendors with demand predictions, dish-level sales trends, and macro suggestions:

  • “Prepare more dal on Monday”

  • “Run an offer on low-sell items at 3 PM”


Vendor AI Assistant (Chat)

Conversational UX layered on top of analytics.


The Impact:

  • Prep errors dropped by 28%

  • Vendor satisfaction score up +17% over 6 months

  • For the first time, kitchens weren’t reacting. They were predicting.

Lorem Ipsum Lorem Ipsum

The Process:
Designing AI people could trust

The Process:
Designing AI people
could trust

AI is powerful, but only if people believe in it. We followed three principles:

  • Progressive Disclosure → start simple, reveal complexity on demand.

  • Explainability → show why a dish was recommended, why an alert was flagged.

  • Human Override → let people correct AI when needed.

This wasn’t black-box magic. It was intelligence designed for partnership.

AI is powerful, but only if people believe in it. We followed three principles:

  • Progressive Disclosure → start simple, reveal complexity on demand.

  • Explainability → show why a dish was recommended, why an alert was flagged.

  • Human Override → let people correct AI when needed.

This wasn’t black-box magic. It was intelligence designed for partnership.

The Results:
Smarter Everyday

The Results:
Smarter Everyday

  • App rating soared: 2.3 → 4.2

  • +13% higher order value from recommendations

  • Repeat orders up 22%

  • Refund tickets down 35%

  • Compliance scores up 21%

  • Vendors reported greater confidence and less waste

  • App rating soared: 2.3 → 4.2

  • +13% higher order value from recommendations

  • Repeat orders up 22%

  • Refund tickets down 35%

  • Compliance scores up 21%

  • Vendors reported greater confidence and less waste

What we built wasn't flashy AI, It was

What we built wasn't flashy AI, It was

Confidence at Scale

Confidence at Scale

Reflection:
Invisible yet Transformative

Reflection:
Invisible yet Transformative

The magic of this project wasn’t the AI models. It was making them invisible.

  • For users, it felt like the app finally understood them.

  • For admins, it felt like decisions made themselves.

  • For vendors, it felt like the kitchen was suddenly smarter.

What we built wasn’t flashy AI. It was confidence at scale.

For me, this was the moment design became more than pixels, it became about shaping intelligence into something people could trust, without even realizing it.

The magic of this project wasn’t the AI models. It was making them invisible.

  • For users, it felt like the app finally understood them.

  • For admins, it felt like decisions made themselves.

  • For vendors, it felt like the kitchen was suddenly smarter.

What we built wasn’t flashy AI. It was confidence at scale.

For me, this was the moment design became more than pixels, it became about shaping intelligence into something people could trust, without even realizing it.

thank you!
keep exploring

thank you!
keep exploring

let's make something
crazyyy!

Find me here

Bringing ideas

to life,

one product

at a time

Made with 🔥

let's make something
crazyyy!

Find me here

Bringing ideas to life,

one product at a time

Made with 🔥