Skip to main content
Tushar Pathak
Let's connect

PortfolioCase study

NP-01Live · mock data

Nuptis

Vendor ops for wedding-planning agencies.

Verification status, work orders, payment milestones and backup coverage in one place — so one vendor no-show stops turning a wedding into a scramble.

Nuptis vendor roster dashboard for a mock agency: active vendors, pending verifications, expiring compliance and backup-tier coverage above a table of vendors with risk tiers.

  • 168

    Figma frames, with an AA audit

    Prototype

  • Day 7

    retired in favour of Velora

    Structural

    still live on mock data

Evidence keyMeasuredStructuralPrototype

The problem

One vendor no-show turns a wedding into a scramble.

Agencies run 15–30+ vendors across 5–7 ceremonies per wedding over spreadsheets and WhatsApp threads — with no structured record of verification, work orders, payments or backups.

How a no-show plays out today
  1. Spreadsheets
  2. Chat threads
  3. A vendor no-shows
  4. No backup on record
  5. A scramble

The insight

Vet vendors by risk, not by habit.

Informed domain reasoning, not fieldwork: the PRD calls Nuptis “a grounding/exploration project”.

  • “Spending three weeks vetting a card printer and two days vetting a fireworks vendor is backwards.”

    — Wedding vendor onboarding process notes

Insight North Star: the % of high-risk work orders with a named backup assigned before the event date.

The product

Onboard by risk, run the event, swap in a backup.

The risk tier set at intake drives the document checklist and whether a backup is required; when a vendor fails, the contingency panel offers only pre-vetted backups.

Nuptis vendor onboarding, step 1 of 5: category and risk tier, with guidance that the risk tier drives the document checklist and backup coverage.
Intake starts with the risk tier
Nuptis procurement board listing mock work orders by wedding, ceremony and vendor, each with a quote and a stage progress bar.
Every work order and its stage
Nuptis “Activate Backup” drawer replacing a mock caterer with pre-vetted backup vendors and an impact summary.
A no-show: pick a pre-vetted backup

Product decisions

Cut hard, and say why.

  • Could have
    Vet every vendor the same way
    Chose
    Risk-tiered verification
    Because
    High-consequence vendors (caterers, fireworks) get scrutiny; a card printer doesn’t need three weeks.
  • Could have
    Build every persona’s full feature set
    Chose
    An explicit cut list, each cut with a reason
    Because
    A fixed nine-day window forces prioritization to be visible, sized by an “effort tracks points” rule.
  • Could have
    Add an AI assistant
    Chose
    A keyword-matching helper; bet on the workflow
    Because
    The bet was the vendor-ops workflow and its data model, not an AI surface the deadline didn’t need.

How it works

The reducer is the API surface.

  1. Agency UIReact single-page app
  2. Context + reducerevery action in one place
  3. localStorageworks offline first
  4. Supabase mirroroptional, four RPCs
  5. Nine-table schema
Nuptis’s local-first data path

Evidence

Designed carefully, measured never — then retired.

Nuptis was evaluated as a design and a build. It is still live on mock data, but the market was a shallow pool, so the same trust problem moved to apparel as Velora.

  • 283

    Figma prototype reactions

    Prototype

    and 62 dual-mode design tokens

  • 9 routes

    swept at 414 and 768 px; 1 bug fixed

    Measured

What isn’t measured

  • No pilot: all three success metrics, the North Star included, are unmeasured.
  • No automated test suite; the README says there is no lint or unit-test script.
  • No usage data of any kind.

What I learned

Say what you know, and how you know it.

  • Name intuition as intuition

    When domain reasoning outruns fieldwork, say so — the PRD called this a grounding project.

  • A North Star can stay dark

    A well-defined metric that never meets a real wedding is still unmeasured.

  • Track the craft too

    The sharpest self-feedback was about prompt engineering and using Claude Code well.

Sources

Everything here is backed by a real artifact.

  • Research
  • PRD
  • Design
  • Readme
  • Evaluation
  • Feedback
  • Post
  • Live data

Evidence behind the Nuptis case study

Private working documents are listed with what they support; only public sources link out.

  1. Week-4 brief

    Research

    Supports: Vendor onboarding stays slow, fragmented and hard to manage across teams

  2. Nuptis PRD

    PRD7 Aug 2026

    Supports: The problem, four roles, the North Star and its unmeasured metrics

  3. Procurement process notes

    Research

    Supports: Risk-tiered verification

  4. PM strategy plan

    PRD

    Supports: “Effort tracks points” prioritization

  5. DESIGN.md (Figma NuptisV2)

    Design6 Aug 2026

    Supports: 168 frames, 283 reactions, 62 tokens, AA audit

  6. README

    Readme

    Supports: The reducer as API; Supabase RPCs; no tests; no AI

  7. App screenshots

    Design

    Supports: Onboarding, procurement and contingency screens

  8. Mobile-sweep note

    Evaluation

    Supports: 9 routes at 414/768 px, 1 bug fixed

  9. Self-feedback

    Feedback

    Supports: Prompt engineering and Claude Code craft

  10. Nine-day series

    Post

    Supports: Retired on day seven for Velora

  11. Nuptis live app (opens in a new tab)

    Live data

    Supports: Still live on mock data

Links and next steps

Solo buildAug 2026Live · mock data

Next

Bhakti Vilas