Applied AI engineer · Founder · CTO
Ross Overstreet
I put AI to work inside the systems businesses run on.
Founder of Rev.io, a billing and payments platform. Co-founder & CTO of Talent Pronto, an AI hiring platform. My team at Overstreet & Co, independent forward-deployed engineers, builds AI agents into the workflows you already run.
IThe movement
The Rev.io years
I founded and ran Rev.io, a billing and fintech SaaS platform for recurring-revenue businesses. Started it in 2002, bootstrapped it through 200+ employees and eight appearances on the Inc. 5000, then exited via an $88 million majority recap to private equity in 2020.
Along the way I held every seat: CEO, CTO, CFO. The software my teams and I built moved more than seventeen billion dollars in payments and invoices for hundreds of companies. For those companies it isn’t a billing tool. It’s the back office: CRM, billing, reporting, and sales channels, the system of record they run the whole business on. The invoices weren’t just processed. Systems I architected and built calculated every line on them: usage, charges, prorations, and the taxes on all of it. Then rendered, printed, and delivered them. Tens of thousands of business users worked in the platform every day, and it served millions of end users.
Systems like these are where I put AI to work now.
- ISales
- IIBilling
- IIICRM
- IVPayments
- VOrders
IIComplications
Track record
-
$17B+
In payments and invoices, calculated and processed by systems I architected and built
-
Hundreds
Of companies run their whole business on it. Their system of record for CRM, billing, reporting, and sales channels.
-
200+
Employees at peak, bootstrapped
-
8×
Inc. 5000 honoree
-
400
Banks I broke into as a hired security consultant
-
Millions
Of end users served by software I architected and built
IIIForward deployed
Hire my team
Every AI vendor now has engineers who will come set up their product. Overstreet & Co is independent. We pick the model and tools that fit the job and your budget, wire them into the systems and workflows you already run, and stay to keep them working as the models change. Small and senior, called in when a project is too important to get wrong.
Most AI projects speed up individual tasks and leave the process around them alone. The days go to the queues and handoffs between steps, so that’s what we go after: we map how the work really moves, cut the steps that shouldn’t exist, write plain code where a rule will do, put agents on the steps that take judgment, and keep people on the ones too risky to get wrong.
Where we start
Finance and back-office work at recurring-revenue companies: quote-to-cash, billing, collections, accounts payable, month-end close, and tax. It’s where I’ve spent most of my career, and where the queues and handoffs cost the most. We take on other work too, and the lists below cover it.
Applied AI
- Agents in your workflows
- Agents that do real work inside your business processes, connected to your data and systems through APIs and MCP servers, with human review where it matters
- Voice and SMS agents
- Agents that answer, call, and text
- Your choice of model
- Claude, OpenAI, or whichever model fits, called directly or run privately in your own cloud when your data has to stay there
- Keeping it working
- Evals that prove the agents do the job, cost tracking on every call, and upgrades to better or cheaper models as they ship
The systems around it
- Billing and payments
- Recurring and usage billing, invoicing, and payment processing, built new or rescued, plus setup and migrations on Rev.io, Chargebee, Recurly, Zuora, or Stripe Billing
- Tax
- Sales, use, and communications tax: audits of how you calculate, charge, and file it; implementation of Avalara, CCH SureTax, or CereTax and their integration with your billing; and cleanup when the numbers don’t line up
- Data and integrations
- Large-scale data migrations off old systems, and connections to CRM, accounting, ERP, tax, and payment platforms
- Rescues
- Stuck and late projects, taken over and seen through to launch
- New builds
- Proofs of concept and new platforms, greenfield or inside the systems you already have, however old
IVFour steps
How it works
Each step is small enough to say yes to. You see how we think before you spend a dollar and know the price before we build. You can stop after any step, and everything we’ve built is handed over to you.
-
01
The call
Free30 minutes
We talk through where you are: work you want AI to take on, a project that’s stuck, or an opportunity you have to move on before it’s gone.
- An honest read on the approach and the risks
- Whether we’re the right fit, and what a Blueprint would cover
- An NDA first, if you want one
-
02
The Blueprint
$2,500Five business days
We map how one workflow really moves and set the numbers we’ll be measured against.
- A working session with me and the principal engineer, and short interviews with the people who do the work
- The real process, read from your systems’ logs where we can get them: every step, loop, and exception, and how often each one happens
- Every step sorted: remove it, automate it with plain code, give it to an agent, or keep a person on it
- A security review: what the agents can reach, what they’re allowed to do, and where a person signs off
- Baseline numbers: cycle time, cost per transaction, exception rate, and how much goes straight through
- A fixed price, a timeline, and the targets the deployment is held to, walked through on a call
Credited in full to your deployment if you start within 30 days. The plan is yours either way.
-
03
The deployment
From$10,000Fixed price, set in your Blueprint
We build it into your workflow and put it live.
- Me and a principal engineer, hands-on in the build
- One workflow live, usually in about two weeks
- Tested against evals agreed in your Blueprint before it goes live
- Security-tested before it goes live, by people who try to break it, with a log of every action the agents take
- Measured against your baseline once it’s live
- A full handoff: code, documentation, accounts, and a walkthrough with your team. You own all of it.
Half to start, and half when it’s live and hitting the targets set in your Blueprint.
-
04
Keep it working
From$2,500A month, per workflow
Optional. We stay on and keep it working as your business and the models change.
- Before-and-after numbers against your baseline at three and six months
- Evals watched, and fixes when something drifts
- Upgrades to better or cheaper models as they ship
- Tuning and small changes as your process changes
- Cancel with 30 days notice, and everything stays yours
Skip it if you’d rather run it yourselves. The handoff at the end of the deployment is complete either way.
Common questions
Do we have to keep paying after launch?
No. When the deployment is done, we hand everything over: the code, the documentation, the accounts, and a walkthrough for your team. Keep it working is there if you want us to stay on, and you can cancel it with 30 days notice.
What if we’re not ready to commit to a deployment?
Start with a Blueprint. For $2,500 you get the real process mapped, baseline numbers, and a fixed price in five business days, credited to the deployment if you go ahead.
Is everything fixed price?
Deployments are. Rescues and open-ended work, where the scope can’t be pinned down up front, are booked by the week: $10,000 for 40 hours of the team’s time, with a plan at the start of each week and deliverables at the end.
How soon can you start?
Most deployments start within a few weeks of the Blueprint. We keep time set aside for emergencies, so urgent work can often start sooner.
How does billing work?
The Blueprint is paid up front. A deployment is half to start and half when it’s live. Keep it working is billed monthly.
Are you tied to an AI vendor?
No. We don’t resell anyone’s platform. We use whichever models and tools fit the work, and we’ll tell you when a cheaper one will do.
How do you handle security?
Agents get only the access their workflow needs, every action they take is logged, and anything risky, like payments, approvals, or data leaving your systems, waits for a person. Before anything goes live we test it the way banks used to hire me to test them: by trying to break in.
Can you work with our own engineers?
Yes. Your engineers are welcome to work alongside us, with my team leading the project, and they make the handoff that much easier.
How do you measure success?
We baseline the workflow in the Blueprint: cycle time, cost per transaction, exception rate, and how much goes straight through. Then we report against that baseline at launch, and again at three and six months.
What if it isn’t working?
You’ll know before it goes live. Every deployment is tested against the evals agreed in your Blueprint, the second half isn’t due until it hits the targets the Blueprint set, and you own everything we’ve built either way.
VAutomatic
AI, put to work
Putting AI to work inside a company takes more than anyone expects. Data has to be organized and reachable, workflows rebuilt around agents, human review designed in, and evals kept current as new models ship. I’ve built AI systems end to end and run them in production, so your team starts from what works.
- Outcomes
- Shorter cycle times, fewer exceptions, and lower cost per transaction, measured against where you started. Your people keep the decisions that matter.
- Latest
- Conversational AI at Talent Pronto, where I’m the primary engineer and have been hands-on in the code for the last two years. It conducts structured interviews on a custom behavior-tree runtime, evaluates candidates against job requirements, and schedules top performers automatically, and it has successfully completed tens of thousands of interviews, turning weeks of screening into minutes.
- Grading
- LLM-as-judge scoring against custom criteria, with a human review queue
- Quality
- An eval harness that compares prompts and models, tests for failure modes like hallucinations, and cost tracking on every call
- Built
- Conversational agents that interview, qualify, and route; unstructured document parsing; and AI wired into existing systems
- Tools
- Claude and OpenAI APIs, LangChain, Langflow, retrieval-augmented generation (RAG), and Claude Code with custom skills
VIWorkshop
What I’ve built so far
Systems I’ve designed, built, and run in production, from AI agents that have completed tens of thousands of interviews to a SaaS platform of 90 backend services and 50+ TB of live data that hundreds of organizations run as their primary system of record.
01
AI
- Conversational agents that interview, qualify, and route
- A custom behavior-tree runtime for structured interviews
- LLM-as-judge scoring, with a human review queue
- Eval harnesses that compare prompts and models
- Cost tracking on every call
- Unstructured document parsing
- Retrieval-augmented generation
- LangChain and Langflow
02
Billing
- Recurring billing
- Usage billing, on a rating engine that handled tens of billions of events a year
- Sales, use, and communications tax
- Bill rendering, tens of millions of PDFs a year
03
Payments
- Credit card and ACH processing
- Payment facilitation
- 50+ gateway integrations
- IVR bill pay and collections
04
Business systems
- CRM
- Workflow
- Two-way SMS and voice systems, with consent compliance
- Transactional email
05
Integration and data
- CRM integrations
- Marketing system integrations
- Accounting and ERP integrations
- Financial systems
- Data warehouses
- Large-scale data migrations
- Industry systems for telecom, hiring, and healthcare
06
Security and compliance
- Penetration testing
- SOC 2, HIPAA, PCI, and ISO 27001 compliance
- GDPR and CCPA privacy
- Field-level encryption
- Multi-tenant role-based access control
100+ systems integrated
- Payments
- Telecom
- Accounting and ERP
- Tax
- CRM
- HR and applicant tracking
- Job boards
- Support and service
- Messaging
- Languages
- Deep in Python and .NET. At home in most languages and frameworks.
- Cloud
- AWS, Azure, Google Cloud, DigitalOcean
- Data
- SQL Server, PostgreSQL, MySQL, SQLite, MongoDB, Redis, Elasticsearch/OpenSearch
VIISpecifications
Background
- Caliber
-
Self-taught, starting at eight years old.
The short version. I read every technical book I could get my hands on. Never cared about playing games on computers. I only ever wanted to build things with them. Still true.
- Tested against
-
400 banks, legally.
Before that. In my early twenties I was a security consultant. Banks hired me to break into them: their networks, their branches, sometimes their locks. I got into 400. Every engagement made me better at building systems that don’t break.
- Power reserve
-
I retired at 38. It lasted a few years.
Time with my kids while they were little. Then back to building, for good.
- Provenance
-
Pensacola, Florida, with deep roots in Atlanta.
Currently in Pensacola, Florida, with deep roots and a lot of ties in Atlanta.
- Off the clock
-
Husband, father of four boys.
I shoot photographs, play music, collect guitars, pick locks, and lift heavy things.
VIIIService history
Timeline
| No. | When | What |
|---|---|---|
| 01 | Age 8 | Self-taught. Read every technical book I could get my hands on. |
| 02 | Early 20s | Security consultant. Broke into 400 banks, legally. |
| 03 | Rev.io | Founded it in 2002 and ran it from every seat: CEO, CTO, CFO. Sold a majority to private equity in 2020 and stepped back from the day-to-day in 2021. |
| 04 | Age 38 | A short retirement while my kids were little. Then back to building, for good. |
| 05 | Now | Putting AI to work: at Talent Pronto, an AI hiring platform, and inside other companies with my team at Overstreet & Co. |