Built by people who needed this tool and could not find it
Deep MM exists because corporate bond pricing gaps are not a solvable problem with existing market data vendors. We built the infrastructure that desks need.
Make every corporate bond priceable
Equity markets have continuous price discovery. Exchange-traded derivatives have continuous price discovery. Corporate bonds do not. A bond that last printed three weeks ago and has no exchange listing has, effectively, no live price. The desk that needs to trade or mark it either guesses or waits.
Deep MM's mission is to eliminate that gap. We build model-derived fair value signals that cover the full universe of traded USD corporate bonds, updated continuously from observable market data: trade prints, dealer axes, and cross-asset inputs.
AI reached the capability threshold in 2023
The corporate bond pricing problem has existed for decades. The key change is that modern machine learning, applied to the right combination of TRACE history, dealer axis data, and cross-asset signals, now achieves accuracy that is actually useful for trading decisions. The 12bp median error on bonds printing monthly, measured in our internal beta program against subsequent TRACE prints, is within the bid-ask spread on all but the most liquid IG names. That threshold matters.
We are not the first company to try to use ML on bond pricing. We are building on years of research showing what works and what does not.
Practitioners and engineers
The team combines fixed-income desk experience with applied machine learning engineering. Both sides are necessary to build a signal that works in production.
Nikhil Krishnan
Chief Executive Officer
Fixed-income quantitative research background, with years building pricing tools on the dealer side. The sparse-market fair value problem was not an abstraction to him: it was a daily friction in the workflow. Deep MM is the tool he spent years wishing existed.
Marcus Webb
Chief Technology Officer
Background in large-scale data pipelines and real-time feed systems. Built the ingestion, feature engineering, and model serving architecture that runs the Deep MM signal. His focus is on reliable low-latency delivery for a model that refreshes on a five-minute cycle across thousands of CUSIPs.
Aiko Tanaka
Head of Quantitative Research
Doctoral-level training in quantitative finance and financial econometrics. Her research focus is credit model calibration and signal extraction from fixed-income microstructure data. The liquidity regime framework and cross-asset feature set that drive the Deep MM ensemble are her work. She leads all backtesting and model validation at the firm.
The problem started on a Tuesday in March
Nikhil was running a corporate bond book. A client called with a $25 million inquiry on a bond that had not printed on TRACE in 19 days. The last available data was a stale broker quote from the week prior, before a 15bp spread-widening event in the sector. Guessing at a fair value was not an option. Quoting wide enough to protect against uncertainty meant pricing the client out.
He made the trade, but left the desk that day convinced that a better tool had to be possible. Modern machine learning had solved harder inference problems than "what is the fair spread on a bond given 19 days without a print." The data existed: TRACE history, CDS spreads, dealer axes. What was missing was the engineering and the model. Deep MM is that missing piece.
Join the early-access program
We are onboarding institutional desks now. Request access to evaluate the signal with your own book.