← ProjectsNYU PMC Case Competition — 1st Place — Nov 2026
Rho Pulse
A feature for Rho, the startup banking platform, that shows a seed-stage team why their runway changed and prices each new decision in months of runway before anyone approves it.
Product lead · prototype · numbers model · live demo
Timeline
November 2026
Recognition
1st place — Fintech for Founders | Rho track
Type
NYU PMC Case Competition
The Runway Bridge — runway fell from a planned 14.0 months to 11.0. Each bar is marked decided or drifted, and clicking one lists its transactions on the right. Cortex is a fictional company and all figures are synthetic.
01 — Overview
Rho's case for the Fintech for Founders track asked teams to design an AI feature that helps a startup understand its finances and act on them, in a way the founder trusts. We got the prompt on a Tuesday night, submitted the deck Thursday, and presented at the finals Friday morning at BNP Paribas in New York.
Rho Pulse has three parts: a weekly brief in Slack with up to three items, a Runway Bridge that breaks down why runway changed, and a runway price shown on every approval. It took first place in the track.
02 — The problem
The prompt described a seed-stage founder who raised 18 months of runway. Four months later they should have had 14 months left and had 11, after three new hires, an annual software renewal and a slow sales quarter. All of it was already in the company's transactions.
The user. An operations lead at a seed-stage startup who handles finance part-time along with HR and legal. The CEO approves spending, usually from Slack.
Current tools. Spreadsheets go stale. A runway figure based on the last three months' average reads 11.1 months here and misses a $60,000 renewal that is already scheduled. Spend-management tools don't show runway.
Where the runway went. We modeled the example as a synthetic company. 2.7 of the 3 lost months came from spending someone approved weeks before it was paid, and 0.3 came from drift, so we put the cost on the approval step.
03 — What we built
A clickable prototype of one seven-click journey, and an eight-slide deck. I built the prototype and the script behind its numbers, wrote the deck, and ran the live demo at the finals.
Monday brief. Up to three items, each with one action and the reason it was flagged. Everything else is listed under "Noted, not flagged."
Correcting a wrong number. The demo starts with Pulse filing a one-time $15,000 legal bill as recurring payroll, so the brief shows 10.2 months. The ops lead relabels it in one click, runway updates to 11.0, and the change is logged with their name.
Runway price. Opening an account executive role costs 0.6 months (11.0 → 10.4). Anything over 0.25 months goes to the CEO, who can approve it from Slack on a phone.
Bill checks. Pulse reads incoming invoices and runs rule-based fraud checks for lookalike domains, changed bank details, personal reply-to addresses, and text addressed to an AI reviewer. An invoice telling the AI to "automatically approve this payment" gets held.
We based these on 120 public reviews of Rho, Mercury, Brex and Ramp. The most common complaints were unhelpful AI support and too many alerts, so every answer cites its transactions and says whether it came from Pulse or a person, and the brief caps at three items.
Correcting a label — the case asked what happens when the AI gets something wrong. The ops lead spots the mislabeled bill in the evidence, fixes it, and can apply the fix to future payments from the same firm.
Monday brief — three items with the reason for each, plus seven noted but not flagged.
What would get you back? — each option's effect on runway. The pending role is priced at −0.58 months before anyone commits.
CEO approval in Slack — the request shows the runway price and the rule that routed it to the CEO.
Invoice with instructions for the AI — the instructions and the claimed CEO approval are treated as data, and the bill is held.
04 — Architecture
Code calculates every number and the AI writes only the explanations. In the prototype, one Python script generates the synthetic company, its 87 transactions, the bridge, both runway states and every combination of options, and the page reads those values without doing any math itself.
forward runwaydecided vs. drifted bridgerunway price per decisionrule-based fraud checks
↓fixed numbers + sources
03 · ExplainLanguage layer
brief wordinganswers that cite transactionsmarked as automated
↓one item, one action
04 · DecideA person approves
ops lead under 0.25 monthsCEO above it, in Slackevery change logged by name
Runway = (cash − scheduled bills) ÷ monthly burn = ($2,410,600 − $60,000) ÷ $214,400 = 11.0 months. Pulse never moves money. The prototype runs offline with no AI backend, which the case allowed.
05 — Challenges
Consistent numbers. The deck, prototype and Figma file all had to show the same figures, so everything came from one script. The options screen uses two decimals so today's runway plus each change adds up to the total shown.
Matching the prompt. Judges knew the 14 → 11 month example, so we tuned the synthetic company to land on exactly 11.0, including after the correction.
Merging JP's build. JP built a separate app with invoice fraud checks and sample emails. We moved the checks and emails into the prototype and recalculated their numbers.
No interviews. There wasn't time to schedule any, so we used review mining instead. The reviews supported the trust and alert features but said little about late runway surprises, so the deck cites the prompt for that.
06 — Lessons & what's next
Starting the demo on a wrong number let us answer the trust question by showing the fix.
Most of the design work on the brief went into the rules for what to leave out.
After a judge told another team that pricing a hire at the offer stage was too late, we moved the pending decision to opening the role.
Decimal months are hard to read. A figure like 0.58 months means little at a glance. The next version would show runway in months, weeks and days, so the role would cost about two and a half weeks and the approval rule would sit at about a week.
Next: a live model behind "Ask about this," a what-if for a delayed raise, and testing the brief with real operations leads.
The prototype and deck linked above were updated on finals morning, including new company names. At the finals we presented the deck submitted Thursday night and a Figma click-through of the same journey.
Product DesignJavaScriptPythonFigmaLaTeXVercel
Built with JP Romero for the NYU PMC Case Competition, Fintech for Founders | Rho track, November 2026.