Oravant

Pricing and liquidity for the prediction economy.

Oravant builds pricing and quoting software for regulated event markets. The engine estimates what a contract is worth and is designed to quote both sides of it; the trader interface puts that estimate beside the prices actually on offer. Built and independently verified. Not yet trading.

Join the early-access list How it fits together
  • Kalshi
  • Polymarket
  • Novig
  • ProphetX
  • Sporttrade
  • Crypto.com
  • ForecastEx
  • Gemini
  • PredictIt

The problem

The prediction economy is growing faster than its liquidity.

Regulated exchanges now list event contracts on sports, economics and politics every day.

Liquidity lags listings

New contracts can arrive faster than professional market makers price them, so books often open one-sided or empty. We target contracts and periods where liquidity is thin and pricing is difficult.

Fair value never sits still

A posted lineup, a weather change or a news item moves the true probability. Quoting well takes models that update continuously and stay calibrated.

Inventory is risk

Two-sided quoting means carrying positions. Limits, hedging and auditability have to be built in from the start.

Retail sees a number, not a reason

A price is not an explanation. Traders need calibrated probabilities and the evidence behind them.

What we are building

Simple in front. Institutional in back.

Two halves of one system, built on the same models and the same controls. Each is labelled with its current status.

Front

OravantQX

All sports. One wager button.

Ask a plain question about a game. Get our model's probability, the price on each venue, and the reasoning behind the gap. No jargon, no spreadsheet.

  • Live game state, pitch by pitch, with our model's probability beside the market's
  • Prices side by side across every venue we cover, with one wager button for every market we serve, subject to venue access
  • An AI analyst that answers questions about the game, our datasets and our models, and draws the chart
  • A rebate on every order, win or lose, tiered by volume, once routing launches
  • All sports at launch; the baseball screen below is one example

Status · Early access. Order routing launches once the required registrations, partner agreements and venue approvals are in place.

Oravant will seek registration as an IB upon funding.

Back

Oravant Engine

A market-maker-grade back end.

Pricing, two-sided quoting, inventory and risk in one system, built to quote across regulated venues from one shared position.

  • Fair value estimated several independent ways, then blended and calibrated
  • A quoting policy sized to fill probability and current inventory
  • Hard limits, authorization gates and an emergency stop on every order

Status · In development. No live trading is enabled.

Business model and roadmap

Oravant is building the pricing and liquidity layer for the prediction economy.

FRONT BACK Customer one wager button OravantQX model probability vs price after fees · explained ask the analyst · charts TODAY Oravant routing competitive price after fees and size partner holds funds PLANNED Venues Integrated Integrated In reviewUNRESOLVED PlannedNEXT CFTC-registered US dollars Oravant Engine market making own capital IN DEVELOPMENT Rebates every order · win or lose tiered by volume PLANNED quotes both sides per order Planned: customer orders and desk trading kept separate. Solid = today · dashed = planned
Full description and status

The business we are building: customer orders flow through Oravant to eligible venues, routed for competitive prices after fees and available size, from one wager button; in the back, Oravant Engine makes markets on its own capital; customers earn a rebate on every order, win or lose, tiered by their volume, under a program whose tiers, eligible volume and payment terms will be published before launch; and OravantQX shows where model-estimated probabilities differ from venue prices after fees, and explains those differences.

In the intended model a regulated partner would hold customer funds and carry accounts. Oravant will seek registration as an IB upon funding. Customer order records and the desk's own trading will be kept separate, and the desk will not trade on customers' pending orders. Customer order routing and the per-order rebates are planned and will launch only once the required registrations, partner agreements and venue approvals are in place; access rules differ by venue, so some routes remain unresolved. Today, OravantQX compares prices and builds a wager but does not transmit orders. Clearing, custody, financing and other regulated services remain future possibilities, subject to the required authorizations, partners and operating capabilities.

This site is a preview. It accepts no orders and no funds. Oravant's intermediary registration, partner agreements and venue approvals are not yet in place. Neither profits, the best price on every trade nor fills are guaranteed. OravantQX provides analysis, not advice.

Products

One set of models, two ways to use them.

The models built for our planned desk are the models behind the analytics.

Oravant Engine

Pricing, quoting and risk in one system.

The software for our planned market-making desk. It estimates contract probabilities and is built to quote both sides across regulated venues while keeping inventory inside hard limits.

Stacked ensembleWalk-forward calibrationOffline RL quotingReplayable journal
Try it · simplified quoting ruleTeaching model, not the production policy
Bid55¢
Ask61¢
Spread6¢
Skew−1¢

Half-spread covers the fee plus the model's uncertainty; a long inventory shades both quotes down so the next fill is more likely to reduce the position.

Status
In development. No live trading is enabled.
Built for
Our own desk, on company capital, once live trading is authorized
Partners
Exchanges and liquidity partners that want deeper, tighter books
Partner with us

OravantQX

Calibrated probabilities, explained.

Our trader interface, in early access. OravantQX shows the same models' probabilities beside the venues' prices; explains the gap in plain language; and answers questions about the game, our datasets and our models, with a chart when one helps.

Is the 54¢ home-team contract fairly priced tonight?

58%Oravant fair value
Market 54¢ on the home side, 11¢ wide Why Starter's last five outings, lineup posted 3:12 pm, wind out to left Confidence Calibrated on this season's realized outcomes
Illustrative example · not a recommendation
For
Traders and analysts who want the reasoning, not just the number
Venues
All CFTC-registered designated contract markets. Customer routing depends on each venue's access rules.
Status
Early access. Prices compared across venues today. Order routing launches once the required registrations, partner agreements and venue approvals are in place
Join the early-access list

Inside OravantQX

One screen, from pitch to price.

The baseball screen shows the interface in early access; all-sports coverage is planned for launch. The game data on this screen is illustrative, not a live feed, and the next-event cards are a simplified view of the full outcome set.

OravantQX baseball screen: game header with at-bat and pitcher, point-in-time explanation, next-event probabilities, Kalshi odds over time, moneyline prices and an open order ticket
OravantQX baseball · early access · illustrative data · next-event cards abbreviated
Live game state

Score, count, runners, and each pitch's speed, type and location.

Game inputs

Pitcher velocity trend, workload and rates; park, wind, humidity, temperature and altitude; series, inning and runners.

Point-in-time explanation

What just moved the probability, in one sentence, with the factors ranked.

Odds over time

Our model's win probability beside the venues' prices, historical and expected over the next half hour.

Prices across venues

Moneyline, total and over/under from every venue we cover, beside our odds, with one wager button.

Next-event probabilities

Illustrative next-event probability cards shown alongside example venue prices.

Ask the analyst

Questions about the game, our datasets or our models, answered in plain language with a chart when one helps.

How it works

From raw data to a resting quote.

Each stage is a separate component with its own contract, verified independently against outside references before anything reaches a market.

Ingest

Official play-by-play, lineups, weather, air quality and news from primary sources, archived with content hashes and provenance.

Point-in-time features

Every input is stamped with the moment it became knowable. Nothing downstream may see the future.

Model arms

Several independent methods estimate outcome probabilities separately, so no single model carries the price on its own.

Calibrate and stack

Those estimates are blended and calibrated against outcomes the model never saw, so a stated probability can be checked against what actually happened.

Quote

Bid and ask are set around fair value and sized to fill probability and inventory, learned from replayed market history rather than live capital.

Execute and reconcile

Orders can reach regulated venues only through authorization and risk gates, and live trading stays off until it is explicitly enabled. When live trading is authorized, every fill will reconcile to a replayable journal.

Accelerated computing

Built to run on GPUs.

Our heaviest workloads are model fitting, simulation, reinforcement learning and inference at scale. Each one is designed around GPU acceleration. The table shows the intended stack and our hardware policy; it is not a record of completed runs.

Production training runs on approved hardware only.Every production fitting entry point validates its hardware before it touches data, and fails closed rather than falling back. Our scaling rule is to profile on one GPU and grow toward eight only from measured throughput, memory and cost.
WorkloadStackHardware policy

Model fitting

Several probability estimators, fit independently and combined.

Probability estimatorsApproved GPUs required

Learned quoting policy

Quoting behaviour is learned from replayed market history, never from live capital.

GPU runtimeApproved GPUs required, fail-closed

Batched state simulation

Large numbers of game simulations run in parallel to turn event probabilities into full price surfaces.

GPU runtimeGPU

Feature pipelines

A multi-year historical corpus joined into point-in-time training sets, with cryptographic receipts on every source and build output.

Columnar data pipelineGPU + CPU

AI analyst

OravantQX's analyst answers questions about the game, our datasets and our models, explains the models' output in plain language and draws charts on request, through a hosted large language model. Self-hosted GPU inference is on the roadmap.

Large language model · chartingHosted model in early access

Built for regulated markets

Fail closed, prove everything.

The controls are not a layer on top of the system. They are the system.

Regulated, dollar-settled venues

Every venue OravantQX covers is a CFTC-registered designated contract market where contracts settle in US dollars. The engine is built for those venues and nowhere else.

Statuses, stated

Prototype, in-development, simulated, paper and live capabilities are labelled separately on this site. Nothing here is live performance.

No look-ahead

Models and policies see only what was knowable at the moment of each decision, enforced by how the data is partitioned rather than by convention.

Reviewed before merge

Every change is reviewed independently before it merges, and design approval requires human sign-off.

Fail closed by default

Authorization, data-validity and risk gates refuse rather than guess. Live trading stays off without explicit authorization.

Replayable end to end

Designed so that every price, quote, order and fill traces back to its inputs and can be replayed exactly.

Team

Founders who write and review the code.

An engineering-led company. The people below build the models, the desk and the product.

Michael Smith

Michael Smith

Co-founder & CEO

LinkedIn

25+ years in engineering, including executive leadership in engineering and AI automation programs at global industrial companies. Leads commercialization and venues.

Richard Ferri

Richard Ferri

Co-founder & Chief AI Officer

LinkedIn

Built Oravant’s pricing engine and leads pricing and quoting. Has been building sports-betting models and software since 2021.

Zak Stallings

Zak Stallings

Co-founder & Chief Product Officer

LinkedIn

Former engineering manager at EarnUp. Built HitParade’s real-time sports-arbitrage prototype. Now leads Oravant’s subscription app, the customer side of the engine.

Company

An AI company for regulated event markets.

Oravant builds and operates its own technology: pricing and quoting software for regulated event markets, built to partner with exchanges and liquidity providers and to trade the company's own capital once live trading is authorized.

Sector
Financial services · AI pricing and liquidity for regulated event markets
What we are
A software company. This site offers no brokerage, exchange or custody services.
Location
United States
Products
Oravant Engine (market making, in development) · OravantQX (trader interface, early access; all-sports coverage planned for launch)
Venues
CFTC-registered designated contract markets
Status
Software in development. No live trading is enabled.
Compute
GPU-accelerated; approved hardware required for production training
Contact
hello@oravant.ai

Get in touch

Let's make the market tighter.

For the OravantQX early-access list, exchange and liquidity partnerships, and investor conversations.

hello@oravant.ai Email us
Traders and analystsJoin the OravantQX early-access list. Exchanges and liquidity partnersDeeper, tighter books on your venue. InvestorsAsk for the deck.