Desk-native AI · Mid-market capital markets
Scoped AI tools for mid-market broker-dealers. Axe distribution, spec-pool stories, relative value at scale. Built inside your compliance, on your infrastructure, and yours to keep.
01 / Approach
A tool earns its spot when it gets three things right. These are the three I build around.
It has to work against real inventory, real accounts, and a salesperson with thirty minutes before the open. I build for the flow the desk actually runs at, and carrying it into that flow is part of the work, not an afterthought.
Your data lives in Bloomberg, the OMS, IDC feeds, and a lot of spreadsheets. The data layer is most of the real work, so I treat it as the starting condition rather than an assumption.
It has to tell a low-loan-balance story from a Ginnie project loan, price a pool against TBA, and read a prepay lockout. The product knowledge is built in, so the tool works inside how you trade.
02 / What I build
Each tool is scoped to a single workflow your desk already runs, and each is built to pay for itself inside the quarter.
Turn the axe sheet and current inventory into targeted, story-ready outreach: which accounts, which bonds, and the angle each buyer actually responds to, drawn from how they have traded with you before.
Generate the prepayment narrative and the comparative pay-up analysis to position specified pools against TBA, matched to the buyers who pay for low loan balance, geography, or seasoning.
Automate the comparative and swap analysis your strategists hand-build for depository and institutional clients: the work that ties up a strategist in Excel for half a day per client.
Draft the daily color, client notes, and education-style material in your firm's voice, with a human in the loop on everything client-facing and a full audit trail behind it. Nothing auto-sends.
03 / Built for your reality
I have delivered to the OCC and built risk systems inside a regulated trading floor. Your compliance constraints are the starting design decision here, not a box checked at the end.
Systems run on your infrastructure and your cloud. Your positions, your client data, and your inventory never leave your environment. This is non-negotiable and it is built in from the first line.
Anything that touches a client, a recommendation, or a price has a person between the model and the world. Nothing auto-sends and nothing auto-trades.
Designed for 17a-4 retention and FINRA 2210 communications review, with the audit trail your CCO needs to supervise it under your written procedures.
Suitability and best-execution logic respected, MNPI walls honored, and MSRB-aware if you run a public-finance or muni book.
04 / Who builds it
For seven years I built the real-time pricing, P&L, and risk systems an agency MBS desk ran on at Citi: an $85B book across thirteen trading desks, delivered into an OCC regulatory program. I know what an axe, an RFQ, a spec-pool pay-up, and a DUS prepay penalty are, because I built the systems that handled them.
Then I led data and AI products in production at Disney Streaming scale: petabyte data, sixty million-plus subscribers, a hundred and sixty countries. That is the proof I can actually ship and run real systems, not just demo them.
Applied Agents is where those two halves meet. I take a small number of firms at a time and stay close enough to be accountable for what ships.
05 / How we start
Every engagement starts with a working call to confirm fit before anything is signed. Then you see one tool work on your actual desk before you commit to anything larger.
One scoped tool, one desk, one workflow. We pick the use case with the fastest payback, I build it on your real flow, and you judge it on whether your producers actually use it.
Embedded part-time with your desk and strategy team to build the next tools, keep the existing ones honest, and bring your people up the curve so the capability stays in-house.
A specific system architected and shipped on your infrastructure, with your team trained to run it and documentation written so you are never dependent on me to keep it alive.
On price: fixed scope, fixed number, agreed before we start. A Desk Pilot is scoped like a few weeks of a senior strategist's loaded cost, not a transformation budget. If it cannot pay for itself on one desk, it is the wrong project and I will tell you so on the call.
06 / Straight answers
You do, and that is the point. It runs on your infrastructure, I build it simple, and I document it and train your team so they can keep it alive without me. Maintenance and handoff are written into the scope, not left as your problem after I leave.
No. Everything runs inside your perimeter and your cloud. Positions, inventory, and client data stay where they are. That is a design constraint from the first decision, not something bolted on for the pitch.
It has to go where they already work, and if it adds a click it has failed. Adoption is the deliverable, not the software. The Desk Pilot exists so you find out on a small, cheap scale before anyone commits to more.
Bring them to the call. The systems are built for 17a-4 recordkeeping and FINRA 2210 communications supervision up front, nothing client-facing auto-sends, and there is a full audit trail. Compliance fluency is the differentiator here, not the obstacle.
I built the systems an MBS desk runs on, so I already speak the product. We start with the work on day one, and the tools get built inside how your desk actually trades.
Nothing but the time. The working call exists to confirm there is a real, scoped use case worth doing. If there is not, you will hear that from me, not a proposal.
Start with a working call. We confirm the use case and the fit before anything is signed.
Book a working call →