Open to opportunities

Jenna Webb

Trust & Safety operator who ships products.

10+ years living at the intersection of risk, data, and strategy— turning behavioral signals, regulatory frameworks, and operational KPIs into programs that actually move the numbers. Across email, messaging, fintech, and SaaS, my most recent Staff PM work used cross-functional, data-driven detection frameworks to cut abuse incidents by 71%. Off-hours, I ship full products end to end across four lanes: detection, live entertainment, operational tooling for real businesses, and quantitative trading.

96%
False-negative rate cut on recent T&S engagement
71%
Abuse incidents cut as Staff PM in a Trust & Safety org
350h
Per month of manual review automated on recent T&S engagement
10+
Years in Trust & Safety

About

Operator first,
builder always

I spent the last decade-plus running Trust & Safety and fraud programs at scale — most recently a Staff Product Manager role leading abuse-prevention strategy for a global email + messaging platform, before that a Fair Billing Compliance program at a major retail-financial-services firm, and most recently a consulting engagement designing onboarding-fraud detection for a notification platform. I’m fluent in the operational side: SQL, Snowflake, Looker, Splunk, behavioral analytics, root-cause investigation, KPI design, carrier and cross-functional escalation.

What sets me apart is that I don’t stop at the spec — I build the thing. The projects below are products I designed, built, and shipped end-to-end using modern AI tooling to move at solo-builder speed without losing the rigor I bring from the T&S side.

The combination is rare: an operator who can talk to legal, finance, and carriers, then go write the schema. That’s the role I’m looking for next.

Selected projects

Four lanes, one builder

Detection, entertainment, process improvement, and trading — different domains, same end-to-end ownership. Each project below was designed, scoped, and shipped by me.

Detection & Verification

Trust & Safety tooling

Fraud-detection APIs, domain risk scoring, OSINT verification, and GRC tooling — shaped by ten years of frontline T&S work.

  • Domain Risk API (SDAT v2)

    Shipped

    Commercial fraud-detection API

    The successor to the Config-checker prototype. Multi-tenant SaaS modeled on industry leaders like eHawk: an API-first product where customers automate fraud decisions instead of running them through a human-review console. The initial version of this engine was installed into a prominent email provider’s infrastructure during a fraud-detection consulting engagement, validating the scoring model on real submission traffic. v2 wraps that proven analyzer in a multi-tenant API with a cross-customer fingerprint reputation network and community feedback loop.

    • Initial version installed into a prominent email provider’s infrastructure for live fraud detection
    • Wraps the proven Config-checker analyzer instead of rewriting it
    • API-first: sync POST /submissions today, async-compatible response shape for tomorrow
    • Django 5
    • Postgres
    • Python
    • REST API
    • +1
    Try the live demo→
  • Config Checker (SDAT v1)

    Shipped

    Streamlit prototype that scores domains for fraud risk

    The original Streamlit-based domain-risk analyzer (~400KB Python). Accepts a domain, returns a structured DomainApprovalResult with risk score, recommendation, and an investigator-friendly summary. Still in active use; the engine that the v2 API now wraps.

    • Single-domain risk scoring with a structured, queryable result object
    • Pulls DMARC, RDAP, and threat-intel signals into one report
    • Investigator-friendly summary explaining the score
    • Python
    • Streamlit
    • RDAP
    • DMARC
    • +1
    Try the live demo→
  • Domain Risk Checker

    Shipped

    Domain & URL fraud analyzer that scores the registrable root AND the exact destination separately

    The evolved Config-checker engine, built on one principle: platform trust shouldn't launder a risky page. Every submission is scored twice — the registrable root and the exact submitted URL/path — so a bad tenant on a trusted host (a scam payhip.com/b/<id>, or an Amazon listing hidden behind an a.co short link) can't hide behind a clean root. It resolves URL shorteners and social/search link-wrappers (a.co, bit.ly, l.facebook.com, google.com/url) to analyze where a link actually goes, optionally renders pages in a headless browser to catch client-side-injected ads, and runs a consumer-harm pipeline that detects made-for-advertising ad-stuffing, pop/push-ad networks, fake-virus scareware, browser-lock traps, and crawler cloaking — with a plain-English summary and analyst-only evidence.

    • Dual scoring: independent verdicts for the registrable root and the exact submitted URL/path, so a trusted host can't whitewash a risky tenant
    • Unwraps URL shorteners + social/search link-wrappers (a.co, bit.ly, l.facebook.com, google.com/url) to analyze the real destination
    • Made-for-advertising / ad-density detection, with headless rendering to surface client-side-injected ads
    • Python
    • Streamlit
    • Playwright
    • Docker
    • +2
    Try the live demo→
  • Social Media Verify

    Shipped

    Streamlit OSINT toolkit

    Enter a domain or email address. The tool auto-discovers social media profiles from the company's own site, verifies each one resolves and references the domain, scans reviews across Google/BBB/Trustpilot/Glassdoor/G2/Yelp/Capterra, detects developer signals (GitHub, npm, Product Hunt, app stores), and computes a 0-100 risk score with Low/Medium/High/Critical tier.

    • Social-profile auto-discovery from a company's own website first
    • Review-site coverage across 7 major platforms (Google, BBB, Trustpilot, etc.)
    • Developer/startup signal detection (GitHub, npm, Product Hunt, Crunchbase)
    • Python
    • Streamlit
    • BeautifulSoup
    • WHOIS
    • +1
    Try the live demo→
  • Risk Graph

    Prototype

    Risk as a linked graph, not a spreadsheet

    A working prototype of a governance, risk, and compliance system built as one linked graph — risks, controls, policies, vendors and systems, metrics, incidents, assessments, and actions — instead of the register-plus-spreadsheets setup most teams live in. Residual risk comes from measured control effectiveness, halved for any control untested in 180 days, and every risk is priced across direct monetary, regulatory, reputational, and customer-churn lenses. Churn is the lens most GRC tools skip: exposed customers × churn probability × lifetime value, where the probability starts as an assumption and is updated by every incident that records how many affected customers actually left (a Beta-binomial update, so a 118-customer incident moves it far more than a 5-customer one). Compliance and fraud share the graph: attack vectors map onto a threat library spanning MITRE ATT&CK, a fraud kill chain, and messaging-abuse patterns, and detections carry TPR, FPR, and time to detect, so the vectors × detections matrix is the fraud-side gap analysis. A gap engine treats absence as a finding, and one graph produces three report packs — board, regulator, and customer trust — that can be issued as hash-chained snapshots the database refuses to edit or delete.

    • One linked graph: risks, controls, policies, vendors/systems, KRIs, incidents, assessments, and actions — every page is a view over it
    • Residual scoring from measured control effectiveness, halved for controls untested in 180 days
    • Churn priced in dollars and learned from incidents: exposed customers × calibrated churn probability × LTV
    • Next.js 16
    • React 19
    • TypeScript
    • SQLite (better-sqlite3)
    • +1
    Try the live demo→
  • SmartStart

    Prototype

    A 90-day trust framework for new senders: accounts earn their sending limits through evidence instead of getting them on day one.

    New accounts are where an email platform knows the least and fraud moves the fastest: a bad actor can sign up, upgrade, blast phishing at full capacity, and abandon the account before traditional detection responds. SmartStart replaces that all-or-nothing access with a lifecycle. Signup signals produce a risk score that sets a starting tier — Restricted, Limited, Developing, or Established — and each tier carries its own daily sending limit. From there, trust is earned, not aged into: an account moves up only after enough clean sending days and enough mail actually delivered, so a dormant account never coasts to full access. High bounces step an account down, severe bounces, complaints, or spam-trap hits restrict it on the spot regardless of tier, and a sudden volume spike pauses promotion. At Day 90 the account leaves the framework but keeps its controls. I wrote the product requirements and then built the simulator to pressure-test the policy: seven scenarios, from a healthy startup to a bust-out that builds trust for a month before turning, play out day by day with every decision explained.

    • Signup risk score sets the starting tier and daily sending limit
    • Promotion needs clean sending days and delivered volume — inactivity never counts as trust
    • Step-downs for soft signals, immediate restriction for severe ones, at any tier
    • JavaScript
    • HTML
    • CSS
    • SVG
    Try the live demo→

Live Entertainment

Platforms for venues, performers, and patrons

Bar-game platforms, booking marketplaces, and mobile games built for real venues under Tangled Webb Entertainment.

  • Tangled Webb Bar Platform

    Shipped

    Live bar entertainment platform plus a public marketing site

    A full multi-game platform for venues, running live bar nights every week and still in active beta testing at the venue — real crowds are the test harness. Players join by scanning a QR code, hosts run rounds from a unified console, and admins manage venues, question banks, promotions, performer profiles, and booking flows. Anti-cheat detects when players leave the app mid-question. A shared prize wheel runs across the games with per-prize stock that pulls a prize once it hits zero, a physical-deck mode for hosts who would rather deal real cards, and landing math centralized behind regression tests after it drifted between the wedge and the prize text. Bar Bingo was rebuilt phone-first this summer: server-side auto-call (the old client-side timer died whenever the host backgrounded the tab), themed square rounds, generated trash talk, and proper handling for two people calling bingo at once. For Veterans Day it gained a third look alongside Classic and Halloween: five new winning shapes (chevron, dog tag, poppy, service star, flag), two 75-square lists, and two ceremonies that take over the TV and every phone — a branch roll call in order of precedence with cues for the host, and a sixty-second moment of remembrance that ends itself on a server-issued clock, because the one person who would remember to end it is holding a microphone. Venues that run their own phone games can also license Tab Out, Triple Down, and the verified trivia bank through service routes that deal a round on the real engine, instead of keeping a copy of the rules that drifts. Tab Out gained sabotage cards and rotating banter. The platform also runs Louie’s weekly hosted game nights. Hangman gives every player their own gallows against the same phrase, and the TV shows only a small hangman per player, never the word or the letters, because anything on the bar TV can be read off it; phrases come from an editable bank and rest 60 days after they’re played. The Corner House Crew is a king-of-the-hill survey-show tournament with a server-enforced buzzer race (the first guarded database update wins, so two simultaneous taps can’t both buzz in), turns that rotate through each team automatically, and boards built from real patron surveys. UNO runs on real decks across up to four independent tables, with the winner spinning the prize wheel on their own phone. Euchre comes two ways: real-deck tables that draw partners blind and swap them after every game to five, and a fully automated phone-and-TV version with bots, going alone, and stick the dealer. The TV displays now survive a crash and hold their screen when the event stream drops. The same Next.js app serves tanglewebb.com — the public marketing site with about, services, schedule, venues, gallery, blog, and testimonials.

    • A dozen game formats share one player join flow and host console, including a Queen of Hearts raffle tracker
    • Player, host, admin, and performer roles each get a tailored interface
    • Shared prize wheel: per-prize stock with auto-removal at zero, physical-deck mode, landing math under regression test
    • Next.js 16
    • React 19
    • TypeScript
    • Turso (libsql)
    • +2
    Try the live Open Mic demo→
  • Louie’s Phone Games

    Shipped

    An eight-game arcade on customers’ own phones

    A games arcade for Louie’s Corner House running off the Tangled Webb platform: trivia, Shut Up & Roll, Tab Out, Triple Down, a flick-to-throw Beer Pong, and three newer skill games — Perfect Pour (hold the tap, let go on the line), Last Call (a reaction test), and Dead Ringer (rings at a swinging hook in a crosswind). Every game plays against the house or head-to-head against another customer through a challenge flow. The index is an arcade: each game gets a lit cabinet screen playing its own mechanic on a loop in inline SVG, so a name like “Dead Ringer” explains itself before anyone taps it. Play is gated behind proof you are in the bar — the ordering app’s geofence, or a code read out by the bartender that is derived rather than stored, changes daily, rolls at 4am because the bar is open past midnight, and skips characters like O/0 and S/5 that get misheard across a loud room. Leaderboards sit deliberately outside that gate, with a chase line (“141 more takes 4th”) because a gap is a target and a rank is just a fact. Every skill game is scored on the server, never the handset: a score computed on a phone is a claim, not an observation, so the client reports events, the server scores and clamps the impossible cases, and those games rank on the leaderboard but can never earn a spin of the prize wheel. Confetti fires only for something actually won — never for simply finishing.

    • Eight games, solo or head-to-head: trivia, Shut Up & Roll, Tab Out, Triple Down, Beer Pong Solo, Perfect Pour, Last Call, Dead Ringer
    • Arcade index with a live animated cabinet per game — the rules explain themselves before anyone taps
    • At-bar gate: geofence, or a daily bartender code derived from the date (no scheduler, rolls at 4am, DST-tested)
    • Next.js 16
    • React 19
    • TypeScript
    • Turso (libsql)
    • +1
    See the live leaderboards→
  • GigHive

    Beta

    Mobile-first booking marketplace where bars and venues book live entertainment

    Two-sided marketplace for live entertainment. Performers publish availability and get booked; bars find acts (especially last-minute) by act type, radius, and region. Includes urgent bookings, fill-in subs, an equipment marketplace, classifieds, and a free public events feed with a follow system.

    • 10 act types, location matching by radius and region
    • $5/month subscription with beta-tester bypass; Stripe handles billing
    • Urgent / fill-in board for last-minute coverage
    • Expo (React Native)
    • Supabase
    • Stripe
    • TypeScript
    View the pitch deck→
  • 2 Drink Memories

    Prototype

    Bar-friendly photo-hunt game

    First title from Fly Trapp Games (a sub-label of Tangled Webb Entertainment). Token-based plays via in-app purchase, image pairs generated through Gemini 2.5 Flash, hotspot hit-detection on tap. Designed for the bar context — short rounds, easy to pick up, hard to put down after two drinks.

    • AI-generated image-pair pipeline with hotspot coordinates
    • Token wallet (10/$1.99, 30/$4.99, 100/$12.99) wired for IAP
    • Tappable hotspot detection with miss-marker feedback
    • Expo SDK 54
    • TypeScript
    • Gemini 2.5 Flash
    • Supabase (planned)
    Play the web build→

Process Improvement

Operational tooling for real businesses

Workflow and customer-facing apps for working venues — replacing phone calls, paper, and interruptions with queues, alerts, and self-service the staff run themselves.

  • Louie’s Phone Ordering

    Shipped

    Phone ordering for a real working bar

    Built for Louie’s Corner House (Buckeye Lake, OH). It started as patio ordering, to stop the patio-to-bar-phone relay that kept pulling bartenders off service, and is now how customers order anywhere in the bar: patio tables have QR codes that bake the seat into the link, and inside orders are picked up at the bar when the phone buzzes. Customers browse a real menu — 300+ items with priced variants, mixers, flavours and bartender-only recipes — order in a tap, edit while they're still in line, message the bar, and get a chime-and-vibrate alert. Payment happens in the app through PayPal Checkout without leaving the page, and a Pay now order reaches the bartender only once the money has actually moved. Every amount sent to PayPal is read from the bar's own rows, never from the phone. Starting a tab no longer means leaving a card behind the bar: the app places a $50 authorization hold, adds another hold when the tab outgrows it, charges the real total at close and releases the rest. A cancelled or declined order is refunded the way it was paid, at the moment it's cancelled, so nobody has to remember. Buy a Round lets a customer treat the room, name up to eight friends with a drink or a dollar amount each, or pay it forward: drinks for nobody in particular that go up on a Wall of Drinks, where anyone in the bar can claim one (one per phone a night, enforced by a unique index rather than a check that two taps could race). Staff approve every round before anything is announced, and then the bell rings on the buyer's phone, the bar screen, and a Wall of Rounds on the TV. Behind it: a specials engine that runs happy hour and one-off bartender specials and recalculates every price server-side at submit; role-based staff PINs; GPS geofencing with a gentle fallback for phones that won't share location; three service modes from open through pickup-only to closed; attributed customer blocking; and a loyalty card that rewards ten different days rather than ten visits or ten dollars, so it can't be farmed by buying four rounds in one sitting. A full card is redeemed by the customer spinning the prize wheel themselves.

    • Full menu with priced variants, option groups and server-authoritative pricing — the phone never decides what anything costs
    • Scheduled specials engine: happy hour by day and hour, one-off bartender specials, scoped to sections or hand-picked products
    • Order from anywhere in the bar: patio tables by QR code, inside orders picked up on a buzz
    • Next.js 16
    • React 19
    • TypeScript
    • Supabase (Postgres)
    • +2
    Play with the working demo→
  • Louie’s Bar Screen & Back Office

    Shipped

    The staff side of the ordering app

    What started as a queue for patio orders became the tool the bar staff run their shift from. Phone orders, paid or on a tab, land in one queue; Buy a Round requests land in another, where a bartender approves one (ringing the bell on the TV) or declines it, which refunds it automatically. Payments made through the Venmo or PayPal links, which the app can't see land, carry a short code like LCH-7F3A so the bartender can find “Sarah, $14” in the Venmo app on a busy night; PayPal Checkout payments are captured and refunded by the app itself. Quick-pick buttons fill from what's actually selling but can be pinned, so a dollar-domestic night gets the domestics up front. Queen of Hearts tickets reserved on the website show up in two piles, still owed and paid-and-ready-to-print, and the bar can see why someone is blocked from reserving and text them from the bar phone. Bartenders take orders at a table on their own phone — set the table once, add each person, and everyone gets their own order and their own tab. Tabs can be split by order (never by dividing a total, because half of a round somebody bought isn’t anybody’s half) or combined; “their usual” turns “another Coors” into one tap; family pricing is applied by the person ringing it rather than as a special anyone standing there would get; and tips written on the slip after a card runs go into a pile to fill in later instead of holding a screen open per slip. A card surcharge was replaced with a cash discount after working through the rules — a surcharge is capped, can’t touch debit, and needs registration, while a discount has none of those problems and reads as a favour rather than a penalty. The night ends at 5am, not midnight, so a 1:30am tab lands on the right day; the drawer is counted against what was rung up as cash, and paid-outs are logged as they happen and voided rather than deleted. Scheduling treats publishing as a step so a late week is visible, totals everyone’s hours in the open, and records every change made after publishing. A time clock and an hours report put scheduled and worked side by side. For outdoor events, a cooler bar sells in one tap with no tab or customer, copies the price onto each sale, and settles owner-stocked gear separately from the bar’s takings. End-of-night reporting covers settlement, comps, tips, top sellers, and event close-out, and staff roles run from bartender through admin to owner.

    • Order and round queues: approve a round to ring the bell on the TV, decline it to refund it automatically
    • Payment matching codes for Venmo/PayPal links; PayPal Checkout captured, held, and refunded by the app
    • Queen of Hearts reservations in owed and ready-to-print piles, with blocklist lookup and text-the-customer
    • Next.js 16
    • React 19
    • TypeScript
    • Supabase (Postgres)
    • +2
  • Louie’s 50th Memory Wall

    Shipped

    Community memory wall for a bar’s 50th anniversary

    Built for Louie’s Corner House (Buckeye Lake, OH, est. 1976) to collect fifty years of customer history before it disappeared into shoeboxes. Patrons upload photos and videos, tag who is in them, add the year and the occasion, and leave comments; stories can be typed or photographed straight off the handwritten wall inside the bar. The wall groups itself by year and event, and a perceptual-hash duplicate detector catches the same photo arriving from four different phones. The team runs it from a dashboard — a review queue that clears in one tap, bulk edit and filtering, name-tag approvals, per-photo comment locks with IP and device capture plus a block list, and a visitor-analytics snapshot. A /tv mode casts the whole thing to the screens at the bar: polaroid frames with Sharpie captions, four-up montages, typed stories as story cards, event flyers and the Queen of Hearts board woven into the rotation, and a rotating ad every ninth set. In production with real patron data since June, and the bar has been playing it nightly ever since.

    • 800+ real patron memories — photo and video uploads, external video embeds, and handwritten stories
    • Perceptual-hash duplicate detection groups the same photo arriving from four different phones
    • Name tagging with approval, year and event grouping, search across names, events, years, and text
    • Next.js 16
    • React 19
    • TypeScript
    • Supabase (Postgres + Storage)
    • +1
    Visit the memory wall→
  • CornerHouse

    Shipped

    The bar’s front door

    The main site for Louie’s Corner House and the hub that ties the other apps together. Everything on it is data, not code: the top nav and the homepage app tiles are driven by rows in a site_links table, so when the next app ships the team adds it themselves and it appears with no deploy. Events carry uploaded flyers, the homepage headlines the current Queen of Hearts jackpot, and a password-gated admin console runs all four surfaces — events, Queen of Hearts, menu options, and site info. The raffle section is the substantial one: a 54-envelope board, live 1-in-N odds, a jackpot trend chart, per-draw history, and admin-editable rules. Logging a draw rolls the jackpot, recomputes the odds, rejects a duplicate envelope, and records the winner automatically when the Queen turns up. Customers reserve raffle tickets online and the bar prints them from its own screen; the default is “I’ll pick numbers at the bar”, because a reservation without card numbers becomes one sticker image with copies set to N on the bar’s handheld label printer instead of one import per number. Pricing lives in one function so the customer’s quote, the admin preview, and the bar screen can never disagree, and the public lookup by reservation code never returns a name or phone. The bar can block a phone number from reserving online, with a plain-spoken message rather than a vague error that would read as a broken app. Each reservation carries two kinds of note kept in separate columns on purpose: a staff note that never leaves the admin screen (“paid half in cash”) and a customer message that appears under the reservation on the customer's own phone (“Card #23 got flipped — come pick a new number”). Reservations flow to the bar screen in the ordering app, sorted into still-owed and paid-and-ready-to-print. Sharing the raffle page produces a live Open Graph card with the current jackpot and cards left. Custom pages carry generic signup sheets — named slots with optional limits — so one primitive covers a 50th-anniversary potluck, a Friendsgiving main-dish sign-up, a chili cook-off entry list (with a QR-coded flyer), a volunteer list, and local businesses offering game-night prizes. Repeating events can have one themed date with its own flyer. The front page shows the games as a physics-driven attract screen (bouncing dice with restitution, a real pendulum, a beer-pong arc solved for its target cup) on a single canvas that stops when scrolled out of view. A first-party analytics tracker shared by all four Louie’s apps feeds a dashboard tab with no IP, name, or phone stored. Admin passwords are self-service from the dashboard — a hash in the database overrides the environment variable and signs every other device out.

    • Nav and homepage tiles are database rows — the team adds new apps without a deploy
    • Queen of Hearts: envelope board, live 1-in-N odds, jackpot trend chart, auto-recorded winner
    • Online raffle reservations printed at the bar — designed around a handheld label printer that takes one image at a time
    • Next.js 16
    • React 19
    • TypeScript
    • Supabase (Postgres)
    • +1
    Visit cornerhouse.vercel.app→

Algorithmic Trading

Signal-mining for markets

Same hypothesis-test-iterate discipline I bring to abuse detection, applied to equities momentum and volatility setups.

  • Gap Up + Fade Screener (v1)

    Shipped

    Pattern-mining predecessor to V9

    The first algorithmic screener I shipped. Configurable thresholds for the premarket gap (>2%) and intraday fade (<-1%), a parallel ThreadPoolExecutor scanning ~90 tickers, and a sortable result table with CSV export. The work that taught me how this signal actually behaves at scale — and what V9 needed to do better.

    • Configurable premarket-gap and intraday-fade thresholds
    • Parallel scan over ~90-ticker small-cap universe; custom watchlist supported
    • Sortable hit table with CSV export
    • Python
    • Streamlit
    • yfinance
    • pandas
    • +1
    Try the live screener→
  • ML Momentum Screener

    Shipped

    Most advanced of the three screeners

    The current production model. Scrapes today's Finviz news per ticker (with a sub-20-minute 'breaking' flag), layers in pre-market and intraday flow signals, and ranks the universe by a composite score. ~1,700 lines of orchestrated screening logic — the iteration on V9 that added external catalyst awareness to the technical signals.

    • Live Finviz news scrape with 'breaking' flag for <20-minute headlines
    • Multi-timeframe momentum (premarket, intraday, 3D, 10D)
    • Composite scoring blending technicals with news catalysts
    • Python
    • Streamlit
    • yfinance
    • BeautifulSoup
    • +2
    Try the live screener→
  • V9 Momentum Breakout Screener

    Shipped

    Real-time multi-signal momentum scanner

    An algorithmic equities screener I built to apply the same signal-mining discipline I used in trust & safety to financial markets. The V9 model fuses six independent technical signals into a single conviction score, ranks the universe live (alphabetical, randomized, or live-volume-ranked), and auto-refreshes every 60 seconds with audio alerts on watchlist hits. Powers the manual side of an end-to-end trading pipeline that includes backtesting, risk controls, and live execution.

    • Six-signal model: EMA10, RSI(7), 3D & 10D momentum, 10D relative volume, VWAP + order flow
    • Scans up to 2,000 tickers with a thread pool; 60-second auto-refresh
    • Three universe modes including live volume-ranked construction
    • Python
    • Streamlit
    • yfinance
    • pandas
    • +2
    Try the live screener→

Experience

Career highlights

Full CV available on request.

Independent Security & Risk Consultant

Notification-platform engagement

Jan 2026 – Apr 2026

★ Built a domain-risk tool that closed a 68% vendor gap
  • Used signal analysis to surface that the incumbent third-party vendor was only catching ~32% of fraudsters — the other ~68% were getting through hijacked-domain attacks the vendor wasn't built to detect.
  • Built a domain-risk analyzer (SDAT — see the Domain Risk API project below) from scratch to target that exact gap, then partnered with Engineering to integrate it into the platform's vetting pipeline — turning a single-vendor pass into a full third-party-plus-supplemental vetting program.
  • Made it ops-configurable on purpose — scoring weights, rules, thresholds, and taxonomies all live in a UI so the compliance team can pivot to new fraud trends without an engineering release cycle.
  • That integration drove a 96% reduction in false-negative rate while holding false-positive rates flat — the signal-health metric I was optimizing for.
  • Same pipeline cut ~350 hours/month of manual review effort and improved enablement time by 90%, with full automation set to push both further.
  • Designed the behavioral risk scoring and identity-verification frameworks behind it — combining device intelligence, document authenticity, network reputation, and behavioral signals.

Staff Product Manager, Trusted Communications

Twilio Inc.

Jun 2022 – Apr 2025

  • Led Trust & Safety strategy for email and messaging abuse prevention protecting millions of users globally.
  • Designed detection and enforcement frameworks that reduced abuse incidents by 71% while minimizing impact to legitimate senders.
  • Built monitoring dashboards and reporting workflows in SQL, Looker, Tableau, Splunk, and Snowflake — used for executive reporting on enforcement, abuse trends, and policy effectiveness.
  • Investigated phishing, spoofing, malicious-link propagation, account compromise, and large-scale spam campaigns.
  • Collaborated cross-functionally with Security, Engineering, Legal, and carrier partners to deploy scalable anti-abuse controls.

Staff Messaging Compliance Program Manager

Twilio Inc.

Nov 2020 – Jun 2022

  • Led compliance and abuse-prevention strategy for A2P messaging ecosystems (10DLC, Short Code, Toll-Free).
  • Built automated enforcement workflows and carrier policy controls that reduced messaging violation rates by 40%.
  • Owned the weekly compliance operating rhythm — triaging carrier developments, escalating critical issues to senior leadership, and driving cross-functional remediation.
  • Defined KPIs and operational dashboards tracking violation frequency, MTTR, escalation trends, and enforcement effectiveness for executive reporting.
  • Improved compliance review throughput by 60% YoY; reduced average incident resolution time by 35%.
  • Developed training materials and internal documentation that scaled regulatory expertise across Product, Engineering, and Operations.

Compliance Specialist

Twilio Inc.

Jun 2020 – Nov 2020

  • Defined the original escalation workflows, monitoring procedures, and enforcement documentation that became the foundation of Twilio's compliance ops.
  • Designed the first-cut tooling and training materials for high-risk-account review — the patterns the team adopted as canonical for suspicious-traffic triage.
  • Served as escalation lead for compliance incidents, customer investigations, and enforcement response coordination — the operating model that informed the Staff PM role that followed.
  • Recognized with the Magic Owl Award for building the compliance department from the ground up.

Fair Billing Compliance Program Manager

Alliance Data (now Bread Financial)

Apr 2012 – Aug 2018

  • Founded and operationalized the Fair Billing compliance program: fraud identification, dispute resolution, ACH/Reg E.
  • Conducted risk assessments and root-cause analyses that reduced billing discrepancies and tightened operational controls.
  • Supported CFPB audits through reporting, operational analysis, and remediation planning.

Also worth noting

  • Quantitative Trading (Apr 2025 – present): Python data pipelines and algorithmic trading strategies across equities, futures, and options — backtesting, risk controls, and live execution.
  • Recognition: Superb Owl Award (Twilio, 10DLC compliance & carrier direct connections), Magic Owl Award (Twilio, building the compliance department from the ground up), President’s Circle nomination (Alliance Data / now Bread Financial, creating the debt-settlement function).
  • Core stack: SQL, Python, Snowflake, Looker, Tableau, Splunk, Jira, event stream analysis, behavioral analytics.

How I work

Operator’s instincts, engineer’s output

End-to-end ownership

Schema, API, mobile, web, billing — I’m comfortable owning the full stack and the product decisions that come with it.

Move fast, with rigor

AI-assisted development lets me ship at solo-builder speed without skipping the thinking — schemas, tests, and architecture still get the time they need.

Real problems

Every project here came from a real pain point — fraud teams drowning in signups, performers hunting gigs, venues looking for better bar nights.

Calm collaboration

I came up in operations. I can talk to legal, finance, and customers — not just engineers — and I write down decisions so the team can move on.

Contact

Let’s talk

I’m open to roles where I can keep building end-to-end and where my range — fraud, payments, marketplaces, live events — is an asset, not a curiosity. Happy to share deeper walkthroughs of any of the projects above.

Full resume available on request.