Spandan Pattanayak
AI-native operator · San Francisco

I decide what
the machine
decides.

I designed the GPT-4 workflow behind Salesforce's incident reports, and an AI deal screener at Draper that clears first-pass triage, so partners spend their hours on the deals worth a meeting. Before that, four years in risk at JPMorgan.

See the work
Reliability Engineering · 2026
ModelConfidence HIGH
GPT-4 draft · P1 · #inc-auth-gateway
A config change cut the auth service's connection limit. The gateway's 503s were downstream of it.
VPReview
Edits the draft, then signs it offApproved

100%of P0/P1 incidents now get an RCA, up from 40%.

Selected work

Where I drew the line

Each one started as slow work someone did by hand.

RCA bot#inc-auth-gateway · 14:09
Confidence HIGH Draft for review · P1
A config change cut the auth service's database connection limit. The gateway's 503s were downstream of it.
Root causeauth-service connection pool exhaustedTriggerconfig deploy, 13:41
13:41Max connections lowered13:44Pool errors on auth-service13:47Gateway returns 503s
Edit & approveApprove

A GPT-4 first draft for every incident

2026 · 12 weeks

RCAs took up to 15 days, and most incidents never got one. I owned the PRD and prompts for a GPT-4 pipeline that drafts them for VP sign-off.

Every P0/P1 now gets an RCA, up from 40%. Delivered 35% faster, with root-cause accuracy up from 60% to 86%.

Read the case · 9 min
Pre-meeting screen · example startupFlag
Founder DNA 30%7.5
Vision, thesis 20%8.0
Technology 20%6.5
Market 15%6.0
Scalability 15%7.0
Bull 15% × 20x × alpha 1.13.30 Base 35% × 3x1.05 Bear 50% × 0.3x0.15 Expected multiple4.5x
5 meeting-only factors left to partnersExample

Triage before the first meeting

2026

Partners were spending their judgment on first-pass triage, so I encoded their criteria into a model that does that pass for them.

Scoring got about 1,000 times cheaper. The five factors only a room can judge stay with the partners.

FeaturedHarmonic's case study on Draper's AI tools
Read the case · 8 min
Pricing model · decision logRisk-as-a-Service
ModelCallWhy
Per calculationRejectedClients ration calls
Share of assetsRejectedFunds won't disclose AUM
Flat subscriptionConsideredHeavy users ride free
License + usageSelectedPredictable base; most clients never see a meter
Entry tier: $80K a year, 50,000 calculations included

Selling JPMorgan's margin engine to hedge funds

2023

JPMorgan's margin engine served only the bank while new rules squeezed hedge funds. I sold it outside, one fund's own portfolio at a time.

3 pilots to 18 clients. $2M contracted in year one. 95% renewed.

“You saved us 18 months and $1.3M.”
COO of a pilot client
Read the case · 7 min
How I work

I start every build with a question from my JPMorgan risk desk days: what do you wish was on your desk at 8am? Then I run people's own data through it before asking for their trust. Eighteen of 21 prospects signed that way.

More impact

The short versions

Six more, from a JPMorgan risk desk to Tim Draper's investment committee. Turn the door to see each one.

Investment committeeCommitted
CompanyHaply Robotics
Sourcingself-sourced
Memo and modelmine
Presented toTim Draper
Outcome$1M
2026

From a cold find to Tim Draper's table

I found Haply Robotics myself, wrote the memo and model, and presented it to Tim Draper at investment committee.

Closed a $1M commitment.

Working group · regulatory deadlineOn time
Salesin the room
Riskin the room
Tradingin the room
Client Servicesin the room
Technologyin the room
Clients onboarded15,000+
Equities risk

A control with no handover

A senior associate left mid-project with a regulatory deadline ahead. I took it over as an analyst, pulled five teams into one working group, and held scope when a Sales SVP pushed a feature that would have cost three weeks.

15,000+ clients onboarded on time.

Leadership scorecardmonthly
Accounts monitored7,000+
Process failures−35%
Downtime−25%
Low-value workstreamscut
Strategy

One forum, one scorecard

I chaired the monthly governance forum across Sales and Trading, Risk and Compliance, set its agenda and OKRs, and built the scorecard leadership used to catch at-risk accounts early.

Failures down 35% across 7,000+ accounts.

Data Science ChampionsSelf-run
Analysts trained30+
Business units4
Program leadrotates yearly
Curriculumreviewed with team heads
Culture

A program that runs without me

I wrote the training material and revision schedule, and set up a lead associate role that rotates each year. Live feedback keeps the content current, and each year ends with a curriculum review with the team heads before the next intern class arrives.

30+ analysts across 4 business units.

Quarterly execution reviewExample
Fundraise plan vs. pipelineon track
LP data roomrefreshed
Board materialsdrafted
2025

Standing in for the partners

I turned the fundraise plan into quarterly execution reviews, ran the LP data room and board materials, and stood in for the partners when they were out.

Ran the cadence through a $30M raise.

Built on my ownevenings
Job-application Chrome extensiontool
Memo generator on MCPtool
SyndicateChaincourse
IndependentOngoing

Things I build for fun

A Chrome extension that fills job applications for me, and a memo generator wired to an MCP knowledge base.

Most of my week runs on Claude Code and Cursor.

1 of 6 · Haply Robotics

Engineer turned operator

Based in
San Francisco
Education
Columbia MBA, 2026
BITS Pilani, engineering
Credentials
Anthropic AI Fluency
Looking for
Founder's office, deployment strategy, AI operations

I build the system a team runs on, one that works without me. JPMorgan taught me to ship controls through teams I didn't run, and during Columbia I took that into venture deals and AI products. Off the clock it's photography and poetry, plus a shelf of about 150 classic novels that runs from Dickens to Dostoevsky.

Where

  • Draper Associates2026
  • Salesforce, Reliability Engineering2026
  • Deep Venture Partners2025
  • Columbia Business School2024–26
  • JPMorgan Chase2020–24
  • BITS PilaniEngineering

Let's talk

If you're deciding what your agents should own, I'd like to help. I'm looking for founder's office, deployment and AI operations roles.

pattanayakspandan@gmail.com
LinkedIn