Bill Baumgarth
AI-Fluent Operator | I build and run revenue systems for sales teams
New York, NY (914) 510-7910 contact@capitalcatalystco.com linkedin.com/in/billb10001 capitalcatalystco.com github.com/Okkay914/SuperArchitect
Summary
On the same 4,622 leads, a leading data provider's records had phone numbers for 8%; my pipeline returned them for 83%. I build and run revenue systems like that for sales teams, across revenue operations (RevOps) and go-to-market (GTM): lead pipelines, data enrichment, outbound, workflow automation and customer relationship management (CRM) intelligence. At Flatiron Realty Capital, a nationwide private lender, I ran business development and marketing and built 22+ systems in 12 months alongside it. I've done the same for clients through my own firm, Capital Catalyst, including 9 institutional meetings booked in one month for a $1B niche credit firm. I design each system, direct AI to write most of the code, and measure results against the data provider's output or the prior period. I started in sales, so I build for the person who has to use it.
Work Experience
Founder & AI-Fluent Operator - Revenue Operations and AI Automation
- Client results: booked 9 institutional meetings in one month for a $1B niche credit firm, with CEOs and CIOs at private-credit firms from $10B to €42B in assets under management (AUM). An executive-search client's recorded meetings rose from 3 a month in Dec 2022 to 39 in Oct 2024, a team result. Alpha Partners has kept me on as its business development (BD) and marketing consultant since my analyst year there ended; I also produced its investor explainer video.
- LLC owner resolution: property lists go in; the people behind the owning LLCs come out, with phone numbers, verified emails and social profiles where found. A deed shows only a shell company, so the pipeline checks 27 state business registries to find the actual owner. A panel of AI models double-checks every match, and every email is verified before delivery (about 28% come back dead).
- Compliance built in: phone lists scrubbed of serial Telephone Consumer Protection Act (TCPA) litigators; leads screened against the NPLA prohibited-borrower list (7,100+ names) and flagged, never deleted; no paid processing past 50 rows without a cost estimate on screen.
- Open source: SuperArchitect (Apr 2025), a Python command-line tool that orchestrates Claude, GPT and Gemini to turn one planning question into a full architecture plan. The zero-budget launch drew 100K+ views and 70+ GitHub stars.
Head Operator (Head of Business Development and Marketing) - Sales Operations, Business Systems and Automation
Nationwide private lender with $2B+ deployed across 2,400+ loans (Flatiron's life-to-date figures). I ran BD, marketing and sales operations, led and trained a two-person marketing team, and was the only person at the 30+ person firm building technical systems: 22+ systems in 12 months, alongside the full-time role. Firm result: loan volume went from $40M when I joined to $120M a year later.
- Beat a leading data provider head to head: on the same 4,622 leads, the provider's records had phone numbers for 8.0%; my pipeline returned them for 83.4%, including 3,505 leads the provider had none for (numbers found, not call-verified), because each record gets its own search instead of one bulk query.
- Opened up a CRM with no API: pulled all 368,683 CRM records out through the screen, then ran a multi-agent AI pass over 253,233 borrower profiles that filled about 1.4M empty fields from half-written notes (where they invest, ground-up or fix-and-flip, last contact). Result: 65.8% of borrowers with a concrete next step and 7,798 dormant borrowers back on outreach, for $20.67 of run cost (about $25,000 at vendor rates).
- Built a lead engine from public records: 26,371 contacts delivered at about 5 cents of API cost each, 18,861 (72%) of them new to the database at delivery. The total includes a county-permit list no vendor sold (4,203 leads with phone numbers). Separately, 56,568 Instagram posts were screened down to 582 developer leads.
- Rebuilt outbound email: 92.63% of 42,438 send requests accepted by receiving servers (Apr to Jun 2026). Sent 68,851 personalized emails to 49,300 recipients from 1,800+ body templates and 8,153 subject lines; 4 in 10 cited a real deal the firm had closed near the recipient, and each new close went automatically to investors active in that market. The system served 21 salespeople.
- Cut the pre-call hunt from minutes to seconds: shipped PersonSearch, a lookup app the sales floor used live on calls. AI agents read signatures, documents and public records and return the owner and a phone number in about 30 seconds, down from 4 to 15 minutes.
- Scored who would pick up: an AI scorer ranked 49,345 stale contacts into call lists. Checked against one rep's logged calls, the top of the list answered 30.8% of the time vs 7.5% at the bottom.
- Moved social from paid boosting to organic video: audited the ad program ($7,741 on 40 boosted posts for 216 direct-message conversations, about $36 each), then stopped boosting and switched to posts and reels made on AI video tools. The first 16 unboosted posts drew 4.6x the views of the 16 before (27,018 vs 5,927; two reels made up 65%), and the best reel's 10,227 views beat the paid program's average of about 2,425 a post. Also rebuilt the brand and the company website.
- Data quality controls: majority-vote AI panels with human-in-the-loop fallback; prompt changes passed regression evals on holdout sets before shipping.
- Regulatory: found a required state registration the firm had never held, and led it through first-time registration and compliance (2026).
Investor Relations / Operations Analyst
- Investor meetings per conference rose from about 5 to 85 at the last one I ran (17x), and the managing partner's qualified-meeting flow at least quadrupled over the year, with institutional limited partners (LPs) across Asia-Pacific and the Americas, for a $1B AUM growth-equity fund.
- Led investor materials: pitch decks, the annual LP meeting presentation and video, and new-fund-launch collateral; ran request for proposal (RFP) and due diligence questionnaire (DDQ) processes end to end.
- First automation work: scripts that resolved CRM data discrepancies, investor-targeting and LP-geography visualizations, and early GPT experiments in the investor relations workflow.
Business Development Associate
- Sourced the company's original $5 million investment and its CTO for an art-focused blockchain platform; recruited more artists than any other salesperson, by a wide margin.
- Ran a 40,000-member community across social channels and the paid ad strategy; won buyers, artists and investors through direct social outreach; began building GPT-3-powered web apps in 2021.
Education
MSc, History of International Relations
Admitted to a department then ranked fourth in the world for history (QS) after only two history courses as an economics undergraduate. Quantitative coursework; dissertation on US–UK relations.
BA in Economics, cum laude (economics GPA 3.8/4.0)
Quantitative concentration: econometrics, statistics and economic modeling.
General Course, visiting year during the BA
Selective one-year program. Earned an A in the European courts and civil-rights law course that LSE law students take a year before graduating.
Skills
- Revenue systems: lead pipelines and waterfall enrichment (Enformion, Trestle, Serper.dev); LLC ownership chains and county property records; entity resolution and deduplication; AI lead scoring; CRM auditing and segmentation at scale (MongoDB-backed); lead extraction from Instagram, LinkedIn, county records and deeds
- Outbound and marketing operations: email deliverability and sender reputation (SendGrid, MillionVerifier); SMS infrastructure (Twilio, A2P 10DLC registration); organic social and short-form video; ad-program audits and budget reallocation
- Experimentation: matched head-to-head comparisons on identical inputs (same leads, two methods); regression evals on holdout sets for prompt changes; scorer checks against logged outcomes (top vs bottom of list); before-and-after comparisons with cost per result; published definitions and caveats for each reported metric; econometrics and statistics coursework
- AI automation and large language model (LLM) systems: production LLM pipelines with model routing (OpenRouter; DeepSeek, Claude, GPT, Gemini, Grok); majority-vote LLM-as-judge panels with human-in-the-loop fallback; agentic retrieval-augmented generation (RAG) that verifies and re-queries; Model Context Protocol (MCP) servers with permission-scoped tools; multi-agent orchestration; structured outputs and tool calling; 93% prompt-cache hit rate; daily agentic coding (Claude Code, OpenAI Codex); using LLMs daily since the GPT-3 Playground (2021)
- Technical / Build: Python (pandas, async); API integrations; idempotent, incremental pipelines; browser automation and scraping (Playwright, Selenium, Apify, Scrapfly); Streamlit, Flask; SQL (PostgreSQL, SQLite), MongoDB, Redis, pgvector; nginx and Vercel hosting
- Compliance: TCPA litigator scrubbing (DNC.com); NPLA prohibited-borrower screening; do-not-contact and consent management; first-time regulatory registration
- Domain: real estate lending (fix-and-flip, ground-up, DSCR, bridge); private equity (PE), venture capital (VC) and private credit (IRR, MOIC, NAV, 13F filings)
Selected Results
From the evidence notes on this site. Each graphic links to how it was measured.
