Sarah Simionescu

Sarah Simionescu is a member of technical staff at Composio and a co-founder of SalesBop, an AI sales-coaching startup acquired in 2025. Her work connects practical product building with the infrastructure those products need: capturing conversations, evaluating sales calls against a team’s practices, and helping agents act across connected applications.

From accessible games to SalesBop

Simionescu began programming with Scratch in fifth grade. An early project, The Hero in the Maze, grew from interviewing a blind neighbor about what she needed from a game. Simionescu built audio cues to guide players through obstacles, locked doors, and hidden objects, then shared the game with students at a school for the blind. Making something that other people could use gave her an early sense of software’s reach. Scratch’s culture of remixing also encouraged her to learn from other developers and share work they could build on. Her account of those early projects connects accessibility with the pleasure of making useful software.

At McMaster University, she pursued computer science alongside research and commercial projects. She was one of six authors of research on utility-preserving face anonymization, published at IEEE INFOCOM in 2024. The collaborative work addressed a difficult balance: concealing identity in facial images while retaining information useful for classification. It combined differential privacy with operations on image features to investigate how privacy protection could coexist with a model’s intended task.

In 2023, she founded SalesBop with Nikos Dritsakos and Jason Huang. Dritsakos had encountered a bottleneck in sales management: useful coaching required someone to listen to lengthy calls and provide feedback. The team built an AI-powered scorecard that evaluated calls against a company’s own sales practices, helping managers identify coaching needs without manually reviewing every recording. Company-specific criteria were central to the product; a useful assessment needed to reflect how a particular team sold.

McMaster’s Forge incubator helped the founders develop a customer-acquisition strategy and connected them with SellWell chief executive Dan Liska. Their collaboration led to SalesBop’s acquisition in 2025. Simionescu graduated from McMaster in 2025 and worked as a software engineer at MongoDB, where she had previously interned. Joining Composio brought her back to startup product building, with work spanning its dashboard and the implementation of a personal agent.

Building the infrastructure behind useful applications

With her university capstone team, Simionescu built MeetingBot, an open-source API for recording meetings. It sends bots into Google Meet, Microsoft Teams, and Zoom calls and makes recordings available to applications through an API. Terraform defines the infrastructure for deployment on AWS, allowing developers to self-host and manage their own meeting data.

MeetingBot tackles a prerequisite for applications such as sales coaching: before software can analyze a conversation, it needs access to the recording. Packaging meeting capture as reusable infrastructure gives developers a starting point for their own applications. Alongside SalesBop, it illustrates two levels of her work—building a product around conversations and building a component that other products can use.

Applications with agents as users

At Composio, Simionescu has worked on the company’s dashboard and built TrustClaw, a personal-agent project. Her work here brings product interfaces, application access, and continuity between tasks into the same engineering problem.

• An interface designed for agents: Simionescu argues that dashboards and application-specific query languages serve as translation layers between a person’s question and the answer they need. Her position on agent interfaces extends beyond connecting applications through MCP: large tool inventories can overwhelm a model, and separate integrations leave the agent to coordinate fragmented capabilities. Tool search helps it find relevant capabilities; execution plans organize how those capabilities work together across applications. The product-design question becomes how to make an application usable by an agent pursuing a task.

• Connected access in TrustClaw: In TrustClaw, Composio brokers OAuth access to applications rather than handing the agent raw application API keys. Tools execute remotely, limiting access to the user’s local machine. These choices put authentication and execution infrastructure behind the agent’s ability to act.

• Continuity between tasks: TrustClaw combines scheduled runs with longer-term memory stored in PostgreSQL using vector search. Web and Telegram interfaces give users ways to interact with it, while scheduling and memory support work that extends beyond a single conversation. Simionescu’s implementation makes those capabilities part of an application people can use.