At Citi, Charles-André Jolly watched post-trade work consume hours that could have been spent with clients. A transaction might begin in a chat or email, then pass through spreadsheets, booking systems, reconciliation teams and compliance logs before it was complete.
Jolly left Citi in 2025 and started Sellara with fellow Citi veteran Spencer Secord, robotics engineer and YC alumnus Ahmad Roumie, and product builder Sebastian Jolly. Their New York startup builds specialized AI agents that extract transaction details from the communications bankers already use and carry that information into the systems where trades are booked, reconciled and reported.
Sellara says it is already live with one of the largest investment banks in the United States. The company has not identified the institution, the workflows in production, the number of users or whether its software submits bookings automatically rather than preparing records for human approval. That distinction matters in regulated institutions, where writing directly into a system of record creates different operational risks from generating a proposed trade ticket.
Turning unstructured messages into transaction records
Banks have spent decades adding software to legacy infrastructure, but employees still move transaction details among chats, emails, spreadsheets and booking platforms by hand. Sellara connects to tools including Outlook, Bloomberg and internal systems, then reads trade recaps, spreadsheets and PDFs to extract fields such as counterparties, instruments, quantities and prices.
The software applies validation rules, routes exceptions for review and retains an audit trail. Sellara's pitch is that models focused on financial workflows can process this material faster and more accurately than general-purpose AI agents while preserving the controls banks require.
The company says an internal reconciliation benchmark reached 98% F1 at roughly three seconds per file. Roumie has also demonstrated a trade-chat model running locally on a CPU, an approach that could appeal to banks seeking lower latency and tighter control over sensitive data. The benchmark remains company-reported, and Sellara has not published enough information about the evaluation dataset, hardware or testing method for outsiders to reproduce it.
Sellara uses a forward-deployed model in which its engineers configure the initial workflow and connect it to a customer's existing systems. It says production deployments can be completed in weeks, though the unnamed bank deployment offers limited public evidence about how the software performs across desks and transaction types.
Founders who handled the work
Jolly worked in credit sales at Citi after graduating from Georgetown University and previously co-founded the education startup Principally. His FINRA BrokerCheck record lists his registration with Citigroup Global Markets from August 2023 through July 2025.
Secord spent six years at Citi across fixed-income, cross-asset and convertible sales. Sellara says he had built internal tools for reconciliation and error reduction before leaving the bank.
Roumie studied robotics at ETH Zurich and previously co-founded rex.fit, a Y Combinator Winter 2023 company that developed computer-vision tools for nutrition tracking. Sebastian Jolly leads product after experience at Capital One and work connected to IBM and Oracle.
Sellara launched in 2025 and joined a16z Speedrun's sixth cohort. It has raised seed funding from investors including a16z Speedrun, Nomad Ventures and IM Ventures.
Competing with established post-trade vendors
Sellara is entering a category with established providers. ipushpull already captures trade information from chat, email and Excel and sends validated records into back-office systems. Duco sells extraction, reconciliation, exception-management and downstream publication software to financial institutions. Banks can also extend incumbent order-management and post-trade systems rather than introduce another vendor into a controlled environment.
Sellara's founders have a credible route into that market because two of them performed the work at Citi and Roumie has already built machine-learning products at a YC-backed company. Sellara's next stage will be expanding that deployment across more desks, transaction types and workflows, giving its specialized models a larger body of live trade data to prove themselves against.