Fixing hiring workflows at scale.
Recruiters lost hours to duplicate candidates and slow shortlisting. I redesigned the hiring module end to end.
~40%faster task completion
60–70%faster shortlisting
Rasoi · Meal planning · 20267 min read · 9 chapters
Designing for the decision, not the plan.
A meal app for Indian home cooks. We were asked for a planner; the research said people don’t plan, they decide. I led the research across 10 designers, then designed the product solo.
- ROLE
- Research lead, then sole designer
- TEAM
- 10 designers in research · design and system solo
- WHAT I DID
- Synthesis, reframe, IA, flows · iOS · 51 screens + design system
- TIMELINE
- 1–2 months · concept, not shipped
Play the problem
← flip the brief, see what research said
- 99survey responses
- 71 → 10notes into insights
- 14decisions logged
- 51screens, designed solo
Plus 8 interviews, 3 run by me. Concept, not shipped: usability testing is next, so no outcome metrics are claimed.
The case in 30 seconds
- PROBLEM
The brief asked for a meal planner. 57% of people don’t plan meals at all.
- INSIGHT
Decide, don’t plan. The real pain is the next meal, two or three times a day.
- WHAT I BUILT
Three dishes you can cook now, a pantry nobody maintains, and a link for the cook.
- NEXT
Not tested yet. A 5-person test of the two riskiest flows comes first.
“Design a culturally relevant meal planning app.”
That was the brief, for health-conscious people in India. It rested on one assumption nobody had written down: that people plan their meals.
- Assumed people plan ahead
- Assumed one user, one cook
- Assumed calorie logging
01 · The brief
Ten designers, five methods, one synthesis.
99 survey responses, 8 interviews, two benchmark matrices across 29 apps, and MyFitnessPal review mining. I assigned the methods, ran three interviews and owned the synthesis.
Assigned the methods
Each designer owned one stream, so nothing overlapped and nothing was missed.
Ran 3 interviews
Sunita, Rohan and Ayesha, so the synthesis wasn’t built only on others’ notes.
Owned the synthesis
71 evidence notes clustered bottom-up, with no themes set in advance.
Three people I interviewed, and what each one needed.
Sunita
Homemaker · cooks for the family
STRUGGLED WITH
Running out of tomatoes mid-cook; setup that took too long
NEEDED
Decide fast, from what’s already at home
Rohan
Fitness professional · has a cook
STRUGGLED WITH
“I ask for a high-protein dish and get something random with paneer.”
NEEDED
A plan the cook can actually follow
Ayesha
Student starter
STRUGGLED WITH
Not knowing what’s in her fridge; apps built for fit people
NEEDED
Simple help, no tracking
02 · Research
Planning wasn’t the problem. Deciding was.
One meal ahead
People plan in their heads. 57% don’t plan at all; 67% of planners do it mentally.
Nobody knows their stock
Tomatoes run out mid-cook, vegetables go bad in the basket. Nobody keeps an inventory.
Localization ≠ translation
54% are confused by ingredient names; 70% want local swaps; 56% still prefer English.
Two people, one plan
The person who plans often isn’t the one who cooks. Intent survives the handover; details don’t.
Rules belong to people
Fasting and diet rules are personal. Only 2 of 29 apps support fasting at all.
Every screen costs
90% never used a planner; 69% of those who tried quit within a week.
03 · What we found
Most people never plan. They decide, two or three times a day.
57%
don’t plan their meals at all
69%
of people who tried a planner quit within a week
2 of 29
apps we benchmarked support fasting
The team started where the brief pointed: a week planner with a calorie tracker. I didn’t argue opinions; I put these numbers in front of the team. We reframed around the daily decision, and planning stayed as an optional tab, not the home.
The new question
How might we make the daily decision fast and confident, in the household’s own kitchen language?
North star
“Aaj kya banau?” answered in under a minute, with a dish you can cook right now.
Planning stays
A week plan is one tap away for the 25% who plan. Today follows it.
04 · The reframe
Six decisions, each with what I rejected.
D1 · Today, not a plan
Three dishes you can cook right now. Rejected: a weekly planner home, a recipe feed.
D2 · Inferred pantry
Stock comes from shopping and cooking, fixed by a tap or a sentence. Rejected: manual inventory, fridge scan.
D3 · Three, with reasons
Each dish says why, like ‘uses your spinach’. Rejected: one ‘best’ dish, an endless feed.
D4 · Five setup steps
Allergies are asked when a dish has one. Rejected: an 18-screen setup quiz.
D10 · A link for the cook
No login, in the cook’s language, with ‘There’s a problem’. Rejected: making the cook install the app.
D11 · Vrat per person
Fasting is a rule on a person; Today adapts. Rejected: a separate ‘vrat recipes’ section.
05 · Decisions
51 screens, six flows, one system.
Every screen traces to a decision on the research board. All of it is built on Sage, my iOS design system: 23 text styles, 18 components, 32 icons.
Concept · iOS · 2026 · 507 component instances
Up close: Today
numbers match the markers on the screen
Onboarding
4 screensToday
4 screensRecipe to plate
4 screensPantry
4 screensCook handover
4 screensThe UI, up close
My first screens looked right. They were built on the wrong job.
The first set had one hero dish on the home screen and no Plan, Kitchen or Family. I tried richer illustration, then new typography. Neither worked, because the structure was wrong.
Structure before style.
What I did instead
Rebuilt every screen from the decision log, then audited my own work like a stranger would.
Traced every screen
Each screen carries a note naming the decision it serves.
Audited the result
19 issues found. Contrast and tap targets were the majors.
“Wasn’t the audit just polish?”
No. It found that search existed but nothing on Today led to it, and that onboarding showed names nobody had entered. Both were fixed.
06 · What I got wrong
Nothing here is validated yet. This is how I’d find out.
- Target
< 1 min
from opening Today to choosing a dish, the north-star job
- Target
Week 1
retention, against the 69% who quit planners in week one
- Target
No call
cook flags resolved in the app, not on WhatsApp
These are pilot measures, not results. First comes a 5-person test of the core loop and the cook handover: three homemakers, one household with a cook, one student.
v1 · test the riskiest
Today, recipe and cook mode, inferred pantry, the cook link, vrat per person.
v2 · for the planners
Week plan, self-building grocery list, leftovers feeding tomorrow.
Later · least evidence
Weekly nutrition nudge and fitness sync. Health had only two participants.
07 · How I’d know
What I’m taking with me.
Question the noun in the brief
‘Meal planner’ hid the real job. The best finding made the brief’s product optional.
Write down what you rejected
The rejected options kept the fridge scan and the calorie diary from creeping back.
Structure before style
Two rounds of visual polish couldn’t fix a home screen built on the wrong job.
Audit your own work
Measuring contrast and tap targets on my own screens found majors I had looked past.
08 · Takeaways
Thanks for reading.
Same instinct, different world: question the brief, then let the evidence decide. That’s also the story of the HRMS work.