15+ pts
3–4 hrs/day
20 hrs → 5 hrs
Accuracy improvement that helped move an AI-assisted workflow into live production
Manual coordination removed from a multi-team workflow
Diagnostic review time during heavy QA weeks
Unique opportunities processed through a personal AI-assisted decision system
01 / HUMAN-IN-THE-LOOP AI
Designing the human system around the model
DoorDash used an AI-assisted matching system to connect merchant items to standardized product records. High-confidence matches could move automatically; low-confidence cases required human review. When a new offshore team initially struggled, the deeper issue wasn’t effort—it was undocumented judgment embedded in the process.
15+ point accuracy improvement
Helped the workflow cross its ~95% graduation threshold and move into live production, with ~98% expected accuracy at live volume.
The model wasn't the only thing that needed to scale.
The human judgment around it did too.
High Confidence
Automated
Potential Match
AI Confidence
Low Confidence
Human Review
Explicit Criteria
QA + Calibration
02 / WORKFLOW AUTOMATION
Turning manual coordination into a system that could run itself
DoorDash’s catalog operation relied on separate internal and vendor-facing trackers, which created hours of manual coordination across request intake, status updates, blockers, escalations, and rework. I progressively turned those trackers into a rules-based workflow system that could validate, route, synchronize, and preserve work automatically.
~3–4 hrs/day removed
Automating the coordination layer eliminated much of the daily work required to maintain trackers, validate requests, route work, reconcile statuses, and manage exceptions—freeing that time for higher-value analysis and process improvement.
I stopped being the connective tissue.
The system became the connective tissue.
V1 / CONNECT
Link internal + vendor trackers
V2 / VALIDATE
Check readiness, permissions, duplicates
V3 / ROUTE
Assign vendor + create timestamps
V4 / HANDLE EXCEPTIONS
Blockers, escalations, revalidation
V5 / PRESERVE + SCALE
Archive rework, protect history, optimize at scale
LinkedIn · Indeed · Built In
SOURCES
Structured opportunity records
INGEST
Missing data · availability · normalization · dedupe
ENRICH + CLEAN
100-point framework · AI-Assisted Score
EVALUATE
HUMAN REVIEW
APPLICATION WORKFLOW
HUMAN APPROVAL
APPLY
03 / AI-ASSISTED DECISION SYSTEM
Building AI around judgment, not instead of it
I built an AI-assisted operating system to process hundreds of job recommendations each week—consolidating fragmented sources, enriching missing data, removing duplicates, evaluating opportunities against an explicit decision framework, and generating application materials. The system handles the repetitive processing while leaving consequential career decisions with me.
opportunities processed
Comparable review time fell from roughly 100 minutes to ~20 minutes per 20 jobs, while tailored application materials went from about 20–30 minutes of manual work to ~3–5 minutes of human review.
Automate the processing.
Preserve the judgment.
MORE SYSTEMS
Selected systems for measurement, data quality, and operational visibility
04 / QUALITY ANALYTICS
Turning QA data into a map of where to look next
100–250 FILES / WEEK
STANDARDIZE
WEIGHT
ERROR DRIVERS
TARGETED INVESTIGATION
~20 hrs → ~5 hrs
Diagnostic review time during heavy QA weeks
05 / DATA QUALITY
Automating the easy decisions. Saving human judgment for the risky ones.
ENTIRE CATALOG
POTENTIAL PAIRS
CONFIRMED DUPLICATES
YES
SAFE MERGE?
NO
AUTO-MERGE
HUMAN REVIEW
8%+ → <2%
Broader duplicate rate reduced to target within weeks
06 / SLA DESIGN
Making “48 hours” mean something
REQUEST RECEIVED
BLOCKED
WORK RESUMES
COMPLETE
0% → 100%
On-time completion reached 100% within five weeks and stayed at 93–100% thereafter
SLA RUNNING
SLA PAUSED
SLA RUNNING
HOW I WORK
Principles I use to turn ambiguity into systems
01
Understand the work before automating it.
I start by learning how the process actually operates, where judgment lives, and which constraints matter.
Make judgment explicit.
If people keep making the same decision, I look for the criteria, assumptions, and edge cases underneath it.
Automate repetition, not accountability.
I use automation to remove repetitive processing while keeping consequential decisions with the right humans.
Measure what helps you act.
A useful metric should do more than describe performance—it should help identify what to investigate or change next.
02
03
04
Let real usage shape the system.
I prefer to build in iterations, use the system, observe the next bottleneck, and improve from there.
05
ABOUT
I’m most useful in the space between teams.
I spent approximately six years at DoorDash working across Merchant Services, New Verticals and Catalog, Vendor Management Operations, and Strategy & Operations.
My formal roles were rooted in operations, but the work increasingly pulled me toward systems—connecting Strategy & Operations, Product and Engineering, Analytics, QA, and frontline vendor teams.
I’m most energized by ambiguous, cross-functional problems where there’s an opportunity to turn a loose process into something clearer, more scalable, and easier for people to operate.
What I’m looking for
Business & Strategy Operations · Product Operations · AI Enablement · Strategic Programs · Systems & Automation
Operating System
Strategy & Operations
Product / Engineering
Analytics
Frontline / Vendor Ops
QA