Greenville, SC · Partnerships & revenue systems

I sell technology
I actually know how to build.

Twenty years in e-commerce revenue — Amazon seller, founding sales hire, partnerships lead. Somewhere in that run I stopped filing requests for the tools I needed and started building them. Everything below came out of a problem I hit inside my own work.

See the workGet in touchgithub.com/rydak81

Live model — not a screenshot

Your pipeline forecast is lying to you about risk

Standard forecasting treats every deal as independent. They aren't — quarter-end pressure, budget cycles, and shared activation capacity make them move together. This is the copula model from my forecasting engine, running in your browser.

Method

How every one of these got built

  1. 01

    Find the friction

    Sitting inside the revenue motion, not adjacent to it. I feel the problem before anyone files a ticket about it.

  2. 02

    Structure the solution

    Define the job to be done, the data required, and where AI does the work versus where a human decides.

  3. 03

    Build it

    Next.js, React, Supabase, Python. Multi-model pipelines with model tier matched to task and cost.

  4. 04

    Ship and iterate

    In front of real users on a cadence — then improve it based on what actually gets used, or shut it down.

Role lens

Reading this for a specific role?

Paste the job description. Claude Haiku reads it against the same data file that renders this page and reorders the shelf for that role. The model selects and orders — every fact on screen still comes from the source file.

Selected work

Five that carry the most weight

Status labels mean what they say. Live is deployed and reachable. Built runs but isn't hosted. Nothing here is inflated to the next tier up.

BuiltThreecolts

Partner ACV Forecast Engine

Pipeline forecasts treat deals as independent events. They are not — quarter-end pressure, macro conditions, and shared activation capacity make them move together. Assuming independence produces a forecast that looks confident and understates real risk in both directions.

+41%
P10–P90 spread widening vs. independence
28% → 34%
Floor-miss probability, corrected
$199K / $182K
Monte Carlo P50 vs. Bayesian P50
+10.3 pts
Top tornado lever
PythonNumPySciPypandasMatplotlibClose CRM API
Read the case study
Runs locally. The interactive above is a browser reimplementation of the same copula model.Read the case study →
LiveFounder

Marketplace Beta

Amazon, Walmart, and marketplace agency operators have no single credible resource tracking what is actually changing in their industry — and that audience is exactly the buyer profile worth reaching.

77+
RSS sources ingested
Next.jsTypeScriptTailwind CSSshadcn/uiSupabase / PostgreSQLVercelVercel CronVercel AI GatewayResendReact EmailClaude SonnetClaude HaikuDALL·E 3
Read the case study
marketplacebeta.comsourceRead the case study →
LiveThreecolts

Amazon Recovery & Profitability Suite

Brands cannot see true per-ASIN profitability, and nobody could credibly size 3P reimbursement recovery without either guessing or overselling it.

2.5× overstated
Correction to initial recovery estimate
−1.30 (R² 0.76)
Rank elasticity of revenue
Gini 0.83 · HHI 1,511
Catalog concentration
20,000
Monte Carlo trials
PythonpandasNumPyMonte Carlo simulationBayesian shrinkageOLS regression
Read the case study
Read the case study →
LiveThreecolts

Weekly Growth Brief Engine

Staying current on marketplace platform changes and competitor moves is roughly a full day of research a week — so in practice it never happens consistently, and go-to-market decisions get made on stale information.

7
Platform surfaces monitored
20
Competitors tracked
ClaudeAutomated web researchPythonopenpyxlStructured HTML reportingScheduled execution
Read the case study
Read the case study →

Everything else

The rest of the shelf

Revenue Systems & Forecasting

LiveThreecolts

Partner Enablement Portal

Partners sign after getting excited about the revenue share, then go quiet and never actually introduce anyone. Time-to-first-referral ran around ninety days, and the standard response — more check-in calls — treats the symptom.

ReactNext.jsRechartsSupabase / PostgreSQLRow-level security
BuiltThreecolts

Prospecting Command Center

Outbound ran across four disconnected systems. Reps spent their day switching tabs and hand-carrying data between tools instead of having conversations.

ReactSupabaseSalesforceApollo APILinkedIn Sales Navigator
LiveVoadera

Business Development CRM & Forecasting Engine

Lead and prospect data lived in spreadsheets disconnected from the company system of record. Outreach status, opportunity potential, and pipeline value were updated by hand and stale by the time anyone looked. Forecasting was an educated guess.

Monday.comERP integrationWorkflow automationMonte Carlo simulationBayesian updating
LiveThreecolts

Partner CRM

A net-new channel program with no system of record, no funnel definition, and no way to see which partners were going cold.

Monday.comWorkflow automationLead scoringExcel
In ProgressThreecolts

Partner Health Dashboard & Slack Workflow

Partner health signals lived across disconnected systems. Nobody saw a partner going quiet until it was a churn conversation instead of a save.

ReactSupabaseSlack APIAutomated workflows
BuiltIndependent

LeadPrompter

LinkedIn Sales Navigator rewards precise Boolean queries and almost nobody writes them well, so reps default to broad searches and work bad lists.

JavaScriptBoolean query generation

AI Products & Pipelines

LiveMarketplace Beta

Article & Topic Classifier

An aggregation engine ingesting at volume is worthless if everything lands in one undifferentiated feed. Relevance has to be decided at ingest, not by the reader.

Claude HaikuSupabaseAutomated pipeline
BuiltMarketplace Beta

PriceScope

Resale pricing is fragmented across eBay, Facebook Marketplace, Craigslist, OfferUp, and Mercari, and no single view tells you what something is actually worth or which listing is genuinely underpriced.

ReactStatistical scoringRecharts
BuiltMarketplace Beta

AgencyForecast

Agencies model their clients’ businesses constantly and their own almost never. The question of which lever actually moves agency profit usually gets answered by instinct.

ReactRechartsWeighted ensemble forecasting
BuiltIndependent

Stellar Advisor Platform

Amazon sellers looking for agency help have no reliable way to find one that fits, and good agencies waste enormous effort on leads that were never a match. Both sides are searching blind.

HTMLJavaScriptVercel

Financial & Acquisition Analysis

BuiltThreecolts

FBA Deal Analyzer Pro

Evaluating an Amazon product opportunity means unit economics, multi-year projection, inventory planning, cash flow, and risk — normally across five disconnected spreadsheets.

Next.jsReactTypeScriptFinancial modeling
LiveThreecolts

3T Recovery Wizard

A recovery estimate on its own is half an answer: a seller sees how large the opportunity could be, but not how likely it is to pay out — or how much to trust the number.

Next.jsReactSupabase / PostgreSQLBayesian updatingError propagation
BuiltBeaconPath Holdings

Universal Business Acquisition Analyzer

Evaluating acquisition targets means normalising inconsistent seller financials, testing valuation against comparable multiples, and modelling whether debt service actually works — repeated on every deal, by hand, in a different spreadsheet each time.

PythonStreamlitpandasFinancial modeling
LiveBeaconPath Holdings

Acquisition Deal Pipeline & Target Database

Acquisition deal flow arrives from brokers, marketplace listings, and direct outreach at once, every target at a different diligence stage with its own financials and seller conversations. A spreadsheet collapses around thirty targets.

NotionAirtableRelational database designPythonStreamlit
AnalysisThreecolts

Carrier & Logistics Spend Analysis

A high-volume shipper suspected carrier overspend but had no way to size it from raw invoice data.

PythonpandasExcelInvoice analysis

Strategy & enablement

  • 1P Market Strategy ReportThreecoltsFindings repositioned the ICP and redirected outbound targeting and messaging.
  • Partner Enablement SystemThreecoltsGave a net-new channel program a repeatable enablement and outreach standard from day one.
  • Company AI All-Hands Case StudyThreecoltsTurned an individual build into an organisational reference point for what applied AI looks like inside a commercial function.

Stack

What I actually work in

Quantitative
Monte Carlo simulationGaussian copula modelingBayesian updating (conjugate priors)OLS regressionBayesian shrinkageSensitivity analysisKalman filtering (graduate coursework)
AI
Claude SonnetClaude HaikuDALL·E 3Multi-model orchestrationModel-tier selection by task and costPrompt engineering and evaluation
Frontend
Next.jsReactTypeScriptTailwind CSSshadcn/uiRecharts
Backend & data
Supabase / PostgreSQLPythonpandasNumPySciPyStreamlitREST APIsScheduled cronopenpyxlpython-pptx
Infrastructure
VercelVercel AI GatewayGitHubResendSlack API
Revenue stack
Monday.com (architected at two companies)SalesforceApolloSales NavigatorDripifyNotionAirtableERP integration
Domain
Amazon 1P / 3P / FBAVendor CentralWalmart MarketplaceTikTok ShopMarketplace recoveryCarrier spendAgency operationsChannel partnerships

The through line

None of these came from a roadmap, a ticket, or an assignment. Each one started the same way: I hit a problem inside my own revenue work and decided the cost of waiting was higher than the cost of building.

That is the actual skill on offer — not React, and not any particular model. It’s the judgment to recognise which problems are worth solving with software, the range to structure and ship the solution, and the commercial instinct to know when a spreadsheet was already the right answer. Twenty years carrying a number is what makes the first and third parts work. The building is what makes them count.

Open to a conversation.

Partnerships and business development, go-to-market engineering, solutions consulting — particularly at companies building for commerce, retail, or logistics, where twenty years of domain knowledge is worth something on day one.

ryandacus@gmail.comlinkedin.com/in/ryandacus-sbcgithub.com/rydak81