Skip to content
Nick Jungmarker

I build data, CRM and AI systems that people can check.

I’m Nick Jungmarker, Data Manager at Environmental Defense Fund with an M.S. in Software Engineering, and a licensed real estate agent in Maryland and DC. Based in Baltimore, working remote.

At EDF I traced a gap between our membership count and the CRM’s to a disabled sync. For my own real estate business I built an AI CRM with a Claude copilot and an MCP server where can send, delete or edit on their own.

Highlighted figures have been checked against their source. Hover, focus or tap one to see where it comes from.

Read the AI CRM case studyDownload resume (PDF)

An AI CRM for my own business

Case study, 2026

Leads from my website, contracts with eight deadlines each, and follow-ups across a two-agent team were slipping through spreadsheets and email. So I built the system I wanted: Claude does the reading and drafting, and a person approves anything that changes a record or reaches a client.

It has , and . These are the rules the code enforces:

  1. Reads run. Writes are proposals.
  2. The AI can’t send, delete or quietly edit.
  3. “Not yours” looks the same as “doesn’t exist.”
  4. Outside text is untrusted.
  5. AI never makes a customer wait.
  6. A person checks every extracted date.
  7. A count always equals the rows behind it.
  8. Every AI action is logged and limited.

Read the case study

Projects

A shipped product, AI agents, and data and GTM systems. Most of the analytics work runs on a synthetic B2B SaaS company I built so it can be shared and scored against a hidden answer key.

Showing 14 projects

Ledgerly CRM MCP

An MCP server that gives sellers and their AI agents governed access to a revenue system: territory-scoped reads, PII masked by role, and changes an agent can propose but never apply.

Write-up ready, code not public yetRoles: AI and agents, GTM Systems

GTM Signal Agent

Finds the accounts that need a seller’s attention, has Claude check the evidence and draft the next step, and puts every draft in a human review queue. Three eval suites score every prompt change.

Write-up ready, code not public yetRoles: AI and agents, GTM Systems

Vurb

A shipped desktop dictation app: hold a key, talk, let go, and the words are typed wherever the cursor is. Speech is transcribed on the user’s own computer, so the audio never leaves it.

Live at vurb.meRoles: AI and agents

Usage Billing and Quote-to-Cash

A dbt and DuckDB warehouse for a usage-priced API business. It prices every metered token against the contract in effect that day, recognizes revenue as it’s consumed, and reconciles usage to invoices to payments.

Write-up ready, code not public yetRoles: Data and analytics, GTM Systems

Lead-to-Pipeline Systems

A Salesforce lead lifecycle and routing design, shipped the way a GTM Systems team would: discovery brief, requirements, SFDX metadata, a replay simulator, UAT tests in CI and release notes.

Write-up ready, code not public yetRoles: GTM Systems

Weekly Business Review Pack

A generator for the weekly metrics review a SaaS leadership team runs on. Every number is defined once in a metrics registry and recomputed a second way before it is shown, and the quarter forecast comes with back-tested error bars.

Write-up ready, code not public yetRoles: Data and analytics, AI and agents

Churn Early Warning

Flags the customers most likely to cancel in the next 90 days, explains each flag in plain English with the numbers behind it, and proves it with an out-of-time back-test and a second company the model never saw.

Write-up ready, code not public yetRoles: Data and analytics, AI and agents

Campaign Lift Lab

Marketing and lifecycle campaigns with a known true effect hidden in a truth file, measured the way a growth analytics team would, then scored on whether the analysis found the truth.

Write-up ready, code not public yetRoles: Data and analytics

Pipeline and Forecast Cockpit

Pipeline analytics plus a quarterly bookings forecast that is back-tested honestly: for every past quarter it rebuilds the pipeline as it looked on day one, forecasts, and compares with what actually closed.

Code on GitHubRoles: Data and analytics, GTM Systems

CRM Dedupe Toolkit

Finds duplicate accounts and contacts in CRM exports, decides which record survives, and produces merge files a Salesforce admin can act on, each with its reasons.

Code on GitHubRoles: Data and analytics, GTM Systems

SaaS Revenue Metrics Warehouse

A dbt and DuckDB warehouse that turns raw contract data into ARR, the MRR bridge, net and gross revenue retention, cohorts and bookings, with every metric defined in writing and backed by tests.

Code on GitHubRoles: Data and analytics, GTM Systems

Support Triage Lab

Suggested routing for support tickets that improves on noisy manual categories and sends anything uncertain to a person, evaluated on accounts and months the model never saw.

Code on GitHubRoles: Data and analytics, AI and agents

Experience

  1. Mar 2022 to Present

    Data Manager

    Environmental Defense Fund / EDF Action

    • Turn requests from fundraising, digital and organizing teams into documented, testable data criteria, and coordinate with vendors and internal technical teams through delivery.
    • Build and validate constituent data workflows across Snowflake, Salesforce NPSP, EveryAction/VAN and Adestra for membership reporting, campaign targeting and data quality.
    • Helped move the organization’s CRM to EveryAction, including data integration, Snowflake reporting and data governance across teams.
    • Traced a ~44K-person gap between engagement-based membership (~112K) and the CRM’s count (~68K) to a disabled Adestra-to-Salesforce sync, quantified the impact and set priorities for the fix.
    • Coordinated identity-resolution QA for a ~1.2M-record deduplication effort, defining primary VAN ID mapping and the checks merge files had to pass.
    • Redesigned acquisition engagement metrics in Hex for 185K+ new supporters. The new definitions capture every action after a supporter’s first one, surfacing 55K+ follow-up actions.
    • Designed weighted phone-scoring rules using evidence from 7 source systems, and caught ~2.7K incorrect vendor-appended phone numbers before they were used in outreach.
    • Delivered targeted outreach lists for field teams in AZ, GA, MI, NV, NC and other states, with a standard of one row per VAN ID.
    • Analyzed a 590K-record voter-file match (Catalist) for patterns in turnout, geography and engagement.
    • Set up ThruText 10DLC texting for an affiliated program, covering opt-in and opt-out handling, privacy policy alignment and launch readiness.

  2. Jun 2021 to Mar 2022

    Manager of Operations Support

    TEKsystems Global Services

    • Managed billing operations for a $50M portfolio and met the department’s quarterly financial targets.
    • Built weekly account and regional performance reports for senior management, and ran team training on data and billing best practices.

  3. Jul 2019 to Jun 2021

    Senior Business Operations Associate

    TEKsystems Global Services

    • Managed accounts receivable over $21M, including enterprise accounts such as T-Mobile, and audited billing data against client requirements.

Alongside, to present

Licensed Real Estate Agent, MD and DC

Long & Foster Real Estate

  • Represent buyers, sellers, landlords and tenants in Baltimore and DC residential transactions, from market analysis and pricing through settlement.
  • Handle offer strategy, contracts, inspection contingencies and repair negotiations, and coordinate with lenders, title companies, inspectors and contractors.
  • Guide first-time buyers through the Maryland Mortgage Program, Baltimore down-payment assistance and HQS rental inspections.
  • Run an independent business on my own AI CRM, website and marketing automation.

Skills and education

AI and automation
Claude API, MCP servers, AI drafting with review and approval, Evals and answer-key scoring, Scheduled jobs (Vercel Cron), REST APIs, Airtable API
Data
SQL, Snowflake, dbt, DuckDB, Python, Data modeling, Identity resolution and deduplication, Data quality and validation, Segmentation, Metric definitions
BI and analytics
Hex, Tableau, Cohort and engagement analytics, Source-system reconciliation, Forecast back-testing, Experiment measurement
CRM and platforms
Salesforce NPSP, Salesforce DX, EveryAction/VAN, Adestra, ThruText, Mobilize, Catalist
Engineering
TypeScript, React, Next.js, Vite, Postgres, Git and GitHub, GitHub Actions, Vercel
Real estate
Bright MLS, Comparative market analysis, Pricing, Contracts and contingencies, Maryland buyer assistance programs, DocuSign
Languages
English, and fluent in Spanish

  • M.S. Information Technology, Software Engineering

    University of Maryland Global Campus

    GPA 3.9, Apr 2024

  • B.S. Agricultural and Resource Economics

    University of Maryland, College Park

    Minor in Global Poverty, May 2019

How I work

I use Claude Code every day as a pair programmer. I write the spec, the business rules and the failure cases. Claude writes most of the code. I review the changes and keep the tests green.

At EDF the habit is the job: turn a vague request into documented, testable criteria, then reconcile the result against the source system. With AI writing the first draft, that checking matters more, so this is how I check its output:

  • Compute it twice.

    Every headline number gets a second, independent computation. In the weekly business review a mismatch withholds the headline instead of footnoting it.

  • Score against an answer key.

    My synthetic datasets carry hidden truth tables the analysis never reads, so dedupe, churn and billing-break detection are scored, not demoed.

  • Break it on purpose.

    A test only counts once I’ve seen it fail. I reverse rules, inject data leaks and corrupt metadata to prove the checks catch them.

  • Prefer real data to fixtures.

    Running against real data caught a duplicate rule that scored 0 for 53 against the answer key. A hand-built fixture never would have.

  • Keep the model to numbers it can back up.

    When Claude drafts a memo, a draft with any figure that isn’t in the fact sheet is rejected.

  • Write down what doesn’t flatter.

    Each project lists its misses, the intervals that came back too narrow and what I couldn’t verify, instead of tuning them away.

Contact

nicholasjungmarker@gmail.com