Work
Three products Rob built, the console he runs his agents through, and a reference build.
Stoke, PomPay Financial, and Tour Time Golf are portfolio companies, not CollectDev clients: Rob Runka partially owns and has helped build Stoke, and PomPay Financial and Tour Time Golf are his own. The Bridge is CollectDev's own tooling, the console the agents on client work are run through. The document annotation service is a reference build, done to a brief in a fixed time budget. They are written up here because they show the kind of software CollectDev builds and the judgement behind it: where agents belong and where they do not, how agents earn trust, money that has to reconcile, and what happens when the legal and payment groundwork comes late. Client work is described on the experience page without naming the client.
Stoke
Agents underwrite inside a lending platform that has to be right.
A gated pipeline of credit agents reads financials, assesses the deal, and checks risk, with every figure grounded, every run scored, and a person making the decision on a double-entry ledger.
Read the case studyPomPay Financial
A storefront that takes stablecoins and keeps no model in the loop.
Crypto checkout, wallet screening, identity verification, and supplier fulfillment run autonomously on rules with an audit trail, because the decisions are regulatory.
Read the case studyTour Time GolfSunset
A tournament platform that worked, and a company that did not.
Registration, live scoring, handicaps, and payouts for amateur golf, shut down over founder agreements, payment processors, and compliance. What it taught CollectDev about launching products.
Read the case studyThe BridgeCollectDev tooling
Agents that run the day. A person signs what leaves.
The console Rob works through. Twelve lanes of agents file, decide, research, draft, review, code, push, and grade themselves every night. One thing stays with a person: approving what goes out under their name. That seam is enforced in the database, and the system keeps score so it can move.
Read the case studyDocument annotation serviceReference build
Upload a document, get a job id, and let an agent pipeline do the reading.
Deterministic extraction, then a cheap model to classify and a stronger one to annotate, each forced through a schema and re-validated before anything is stored. Postgres is the queue. Built in a fixed time budget, with the review passes that found its own bugs written up.
Read the case study