Book a Demo
The World's First Large Event Model (LEM)

The market doesn't
know this exists yet.

A new class of artificial intelligence — built specifically for financial markets — has been invented. Not adapted from language models. Not a better quant model. Something entirely new. And it will define which trading businesses lead the next decade.

Book a Private Demo Why This Changes Everything
+44%
Pricing Accuracy
Investment Grade improvement
+48%
Pricing Accuracy
High Yield improvement
400K+
Trades Evaluated
Out-of-sample, Jul–Sep 2025
Live
Top-5 Global Bank
Top-5 bank, systematic credit desk
Scroll
The P&L Impact

Here's what that's
worth to your desk.

$86M
Annual benefit at 1% market share
Across IG + HY combined
$268M
Annual benefit at league table share
Top-5 global bank equivalent

For systematic desks, every basis point of signal improvement compounds across thousands of trades. The accuracy gap is measurable, cumulative P&L.

$0.0383/side improvement on IG ($15T annual volume) + $0.0748/side on HY ($3.75T annual volume) · 1% market share · vs industry published benchmark

"We've run this for every major dealer. A custom P&L impact estimate with your name on it."

Available on request →
The Paradigm Shift
What Every Firm Has Been Asking

"How do we use AI to cut costs?"

Large Language Models transformed industries built on words. They automate, summarise, and generate text — cost reduction, fewer people doing the same work.

For trading businesses, this is valuable. But it does not grow your trading revenue by a single dollar.

The Real Question

"How do we use AI to grow revenue?"

Financial markets are not built on words. They are built on events. Every trade, every quote, every rate move, every signal firing simultaneously across thousands of instruments.

No language model was ever designed to understand that world. Deep MM built one that is.

Deep MM did not build a better pricing tool. We built the first intelligence ever designed to understand a market.

Every firm is using the same signals — incumbent pricing models built on legacy technology with fewer than 1,000 parameters. These are the industry benchmarks. They define what the market treats as correct.

AXOR runs on 80 million parameters. Legacy models run on fewer than 1,000.

AXOR is the first signal ever independently validated as meaningfully more accurate than those benchmarks. By +44% on investment grade. By +48% on high yield. Evaluated on 400,000+ out-of-sample trades. Already live at a top-5 global bank.

A New Class of AI

A new class of AI
has been invented.

Deep MM's AXOR is the world's first Large Event Model — a new class of AI trained on market events, not language. Not adapted from a language model. Not a better quant model. Built from the ground up to understand what actually moves credit markets.

Large Language Model

Trained on words.

Transforms industries built on text — automates tasks, cuts costs, reduces headcount.

The Old Thinking

Does not grow your trading revenue by a single dollar.

Large Event Model — Deep MM AXOR

Trained on market events.

Processing thousands of signals simultaneously. Every trade, every quote, every rate move. Updating every 15–30 seconds. Seeing patterns no legacy model has the capacity to detect.

80 million parameters. 2,000+ intraday features per inference. The model knows the difference between a $100K offer and a $5M offer — because it was trained to.

Not a tool. Not a data feed. The signal that makes every other model in your stack more accurate.

80M
Parameters in AXOR
vs <1,000 in legacy Kalman-filter models
2,000+
Features per inference
Updating every 15–30 seconds throughout the trading day
19
Percentile outputs per inference
5th through 95th — the full probability distribution
The Key Difference

Not a price. A probability distribution.

While some signals include a confidence score, AXOR delivers the full probability distribution — the 5th through 95th percentile of possible outcomes, conditioned on your specific trade size, side, and market conditions. The market doesn't trade at a single price. Your signal shouldn't either.

Legacy Pricing Signal

Output: 127 bps — A single point estimate. No uncertainty. No size conditioning. The same number whether you're trading $100K or $5M.

AXOR — Large Event Model

Output: Full distribution — The 5th through 95th percentile of spread outcomes, conditioned on your specific trade: $1M notional, offer side, US IG corporate bond. The market doesn't trade at a single price. • Your signal shouldn't either.

This is why AXOR outperforms on large trades. The model is conditioned on trade size and side — a $5M offer produces a materially different distribution than a $100K bid. A confidence score tells you how certain a model is about its single estimate. AXOR's distribution tells you the full range of probable outcomes. Your risk is not the same on every trade — your signal should reflect that.

Proven Accuracy

The numbers are unambiguous.

Validated against the industry's published accuracy benchmarks. $0.049 median price error on IG. $0.082 on HY. Evaluated on 400,000+ out-of-sample trades, Jul–Sep 2025. The first model to demonstrably outperform those benchmarks at scale.1

44
+44%
Pricing Accuracy — Investment Grade

Lower median pricing error across a large sample — more consistently closer to the true market price than any competing signal, across all market conditions including stress.

48
+48%
Pricing Accuracy — High Yield

Superior precision on riskier, lower-rated bonds — exactly where pricing errors are most costly to a trading P&L, and exactly where legacy models degrade most severely.

01
Lower median error

More consistently closer to the true market price across all instruments and market regimes.

02
Resilient in volatility

All models degrade under stress. AXOR degrades significantly less — widening the accuracy gap exactly when stakes are highest.

03
Better on large trades

Where the P&L impact of a wrong signal is greatest — AXOR's size-conditional inference means it was built for this.

04
Precise on risky bonds

Superior accuracy on lower-rated, less liquid bonds — exactly where pricing errors are most costly to a trading P&L.

05
Faster response to macro moves

When rates shift or macro conditions change, AXOR maintains its accuracy advantage. Legacy models degrade most severely under macro stress — exactly when you need your signal most.

Independently validated  ·  Live at a top-5 global bank  ·  The first model to demonstrably outperform the industry benchmarks at scale

Independently Validated
$0.049
Median IG Price Error
vs $0.088 published industry benchmark
$0.082
Median HY Price Error
vs $0.157 published industry benchmark

AXOR's price model was evaluated out-of-sample on 400,000+ trades from July through early September, 2025 — entirely after the model's model training cutoff, ensuring no data leakage. Only trades of notional size $150k or above were included.

Deep MM's median price error of $0.049 on IG and $0.082 on HY represents a 44% and 48% improvement respectively over the industry's published accuracy benchmarks. The gap holds consistently across trade sizes and throughout the trading day — including under volatile market conditions where legacy models degrade most.1

1 Compared to CP+'s published accuracy benchmark. Methodology available upon request. Test sets are independent; CP+ figures sourced from the MarketAxess CP+ Whitepaper (2024). Past performance is not indicative of future results.

Live Proof — Top-5 Global Bank

Added Deep MM as
an ensemble input.
Got 12% more accurate.

Integrated AXOR pricing alongside their existing signals — not as a replacement. The ensemble approach delivered a 12% improvement in MAE accuracy on their mid prices immediately.

Deep MM does not require you to rebuild your model. It makes the model you already have significantly more accurate. Drop it in as an additional input and measure the improvement yourself.

They put Deep MM through a full evaluation on their own trading flow. The results spoke for themselves. They are now a paying client in live production.

+12%
MAE accuracy improvement
On mid prices · Ensemble · IG + HY · Live production
Also In Evaluation

Top-10 bulge bracket institutions currently in active evaluation. The desks that move first hold a signal advantage that is very difficult to close.

How It Works

Built differently.
By design.

AXOR is fed a live stream of real-time data the deep neural network has learned to extract signal from: FINRA TRACE (every US corporate bond trade), real-time treasury prices at all benchmark tenors, and equity market signals including SPY, QQQ, HYG, JNK and LQD — plus S&P credit ratings. Every inference is then conditioned on your specific trade parameters — bond identifier, quantity, side, and ATS indicator — producing a probability distribution unique to your trade.

01

FINRA TRACE Feed

Direct connection to FINRA's TRACE — every US corporate bond trade as it occurs, alongside real-time treasury prices at all benchmark tenors and live ETF market signals.

02

2,000+ Feature Extraction

AXOR extracts over 2,000 features per inference from TRACE prints, treasury benchmarks, SPY, QQQ, HYG, JNK, LQD and S&P ratings — simultaneously, in real time.

03

80M Parameter Inference

The LEM processes all signals through 80 million parameters — conditioned on your trade size, side, and ATS indicator — producing a complete probability distribution.

04

WebSocket Delivery

Subscribe to any bond in your universe. Receive the full distribution — 5th through 95th percentile — via WebSocket, refreshing every 15–30 seconds.

Platform-agnostic. No lock-in.

AXOR is delivered via WebSocket connection — one of the most universal data protocols in modern software. Your quant team can integrate the signal into any existing system or data pipeline without migrating off the tools your desk already uses.

No proprietary platform required. No new terminal. Connect once, subscribe to any bond in your universe, route the signal wherever you need it.

See AXOR Live
US Corporate Bonds~20,000 IG + HY securities covered
Price, Spread, or YTMChoose your label per inference
5th–95th Percentile19-point probability distribution
Size-Conditional$1K to $5MM notional supported
Bid / Offer / DealerFull two-sided market view
Historical InferenceBack to January 1, 2019
Up to 32,000 SubscriptionsPer connection simultaneously
Strategic Vision

Built to scale across
your entire firm.

A Large Event Model requires a real-time trade reporting feed — the same way TRACE powers AXOR today. Any market where events are reported in real time is a market where a LEM can be built.

Live Today
IG Corporate Bonds HY Corporate Bonds
Future Vision
EUR Corporate Bonds GBP Corporate Bonds US Treasuries Municipal Bonds Agency MBS Listed Equities FX Interest Rate Swaps CDS Equity Options Cryptocurrencies

This is not a desk-level tool. It is AI infrastructure for the entire trading enterprise. We begin with US IG and HY corporate bonds — the market where pricing accuracy has the highest immediate P&L impact. From there the architecture scales.

Book a Demo
Integration

Built for systematic trading.

API-First Signal-Ready Zero Integration Risk
Simple API

Real-time prices via WebSocket.

Send CUSIP + notional + side. Bid, dealer, and offer streams update continuously throughout the trading day.

No client data required.

Chance-to-Fill

Fill probability for every bond.

Powers smarter RFQ routing, quote optimization, and execution strategy for systematic desks.

Ensemble-Ready

Drop in alongside your existing signals.

Measure the accuracy lift immediately. Deep MM does not require you to rebuild your model.

No model rebuild required.

No data to share. No model to rebuild. Just a better signal.

AXOR QX — Quoting Engine

Your objective function.
Our distribution.
The optimal quote.

AXOR QX takes AXOR's probability distribution and your desk's own objectives — risk tolerance, market share, client ratings — and derives the mathematically optimal quote price for every RFQ. Automatically. Included at no cost.

Quote Optimizer

Optimal quotes on every RFQ — automatically.

Production-ready reference implementation. Uses the full AXOR probability distribution and your desk's objective function to derive the optimal quote price on every single RFQ.

Sample Code

Yours to own, extend, and build on.

Unfettered perpetual license. No lock-in, no dependency on Deep MM to operate it. Your quants take it, understand it, and make it theirs. A running start — not a black box.

Consulting Hours

We calibrate it to your desk's objectives.

Deep MM quants work with your team on objective function calibration, quote optimization strategy, and backtesting support — using your flow, your history, your goals.

Implementation Support

Integrated into your existing stack.

We work with your engineering team to integrate AXOR QX into your existing infrastructure — whether that's an internal quoting system, a proprietary execution platform, or a custom algorithmic flow.

Your quants know your flow better than anyone.

AXOR QX gives them a running start — not a replacement. Use as much or as little of it as helpful. The objective function is fully configurable: risk tolerance, market share targets, client ratings, long/short bias — all adjustable in real time without retraining the model.

Optional Included at No Cost Perpetual License No Lock-In
Book a Demo to See AXOR QX Live
The Team

Experience at
your service.

Deep MM brings together decades of machine learning at the world's biggest institutions — JPMorgan, Microsoft — with deep credit trading expertise. We built what the big banks couldn't, because we know exactly how they think.

NP
Nathaniel Powell
Founder & CEO

Nathan built JPMorgan's corporate bond pricing model at the world's largest US credit market maker. He brings that institutional knowledge — plus 15+ years of ML and software engineering — to every firm that accesses AXOR.

Executive Director of Machine Learning — JPMorgan Chase
Head of ML R&D — AIM2
Software Development Engineer — Microsoft (Data Science & ML)
View LinkedIn →
BA
Brian Adams
Principal Lead ML Engineer

Brian leads the engineering architecture that makes AXOR production-grade — high-performance, low-latency systems delivering the signal with bank-grade reliability under any market condition, including maximum volatility.

Senior software engineer across high-performance systems
Specialist in real-time data infrastructure
B.S. Computer Science — Brigham Young University
View LinkedIn →
JF
James Finer
Head of Sales

James spent 16+ years at Goldman Sachs trading every corner of the capital structure through every credit cycle since the GFC. At Deep MM he bridges the world most people keep separate — he speaks trader, and he knows exactly what AXOR can do for a desk like yours.

16+ years at Goldman Sachs — HY, IG, loans, CDS, CDX, converts, swaps
$20B GSAM high-yield platform · $2.5B Liberty Harbor hedge fund
$10B+ annualized HY trading volume · full capital structure expertise
View LinkedIn →

Deep MM did not build a better pricing tool.

We built the first intelligence ever designed to understand a market.

Our mission is to build AI analysts that help all credit professionals gain insight, confidence, and empowerment. AXOR is purpose-built for US IG and HY corporate bonds — the market where pricing accuracy has the highest P&L impact, and where legacy models fall furthest short.

We are expending more resources concentrated on US corporate bond pricing than JPMorgan was when Nathaniel built their model. This is all we do. And it shows in the numbers.

The Moment You Are In Right Now

This is the window.
Move first.

Think back to 2021. Before ChatGPT. Before every board was asking about AI strategy. There was a brief window when a small number of firms understood what was coming and moved first. This is that window for trading. The Large Event Model is live. A top-5 competitor is already using it. The rest of the industry will know within 12 to 18 months.

Book a Demo Now
Founder & CEO
Nathaniel Powell
nathan@deepmm.com
Head of Sales
James Finer
sales@deepmm.com
Phone
Deep MM
(801) 623-2785
Private Briefing — Available Now

See AXOR
in action.

The world's first Large Event Model for financial markets is live and delivering results. In a private demo you will see exactly how AXOR works, what it outputs, and what it would mean for your desk — with your bonds, your trade sizes, your use case.

44%
Pricing accuracy improvement
Investment Grade
48%
Pricing accuracy improvement
High Yield
80M
Model parameters
Large Event Model — AXOR
Live
Top-5 global bank
Already deployed
What to Expect
1
A live look at AXOR's output

See the full probability distribution on real bonds — price, spread, or YTM — conditioned on your specific trade size and side.

2
The accuracy results

We walk through the methodology behind our 44% and 48% improvement figures and show you what that means in dollar terms for a desk like yours.

3
Your integration path

WebSocket delivery means you connect once and route the signal into your existing systems — no platform migration, no new terminal.

4
Your bonds, your universe

If you have a sample trade list, bring it. We can run it through AXOR live so you see your instruments, not a generic example.

Or Reach Us Directly
Founder & CEO
Nathaniel Powell
nathan@deepmm.com
Head of Sales
James Finer
sales@deepmm.com
Phone
Deep MM
(801) 623-2785
Schedule a Private Briefing

Pick a time with Nathaniel.

30 minutes. Live AXOR walk-through tailored to your desk. Bring questions, bring bonds.

Loading scheduler