Leeroo

What do they actually do

Leeroo sells a platform of continuously learning AI agents that behave like specialist teammates for data and AI work (data engineers, analysts, ML engineers). Customers can deploy and adapt pre-built agent blueprints such as churn prediction, data quality, customer 360, KPI orchestration, live dashboards, and anomaly detection to their own stack Leeroo agents Leeroo company.

Enterprises can run Leeroo as a dedicated SaaS instance or deploy it inside their VPC/on‑prem. A forward‑deployed engineer can help map data sources, connect tools, and take agents from pilot to production. Agents connect to a customer’s documentation, data, and tooling; user interactions and production traces are fed into a learning engine that updates the agents over time. Early pilots target data/AI teams in regulated industries like banking, finance, and healthcare, and the company is an active YC S25 startup with a small team (listed team size: 3) Leeroo site Founding Engineer role YC page.

Under the hood, Leeroo publishes an LLM orchestration repo and notes a production stack using vLLM on AWS EC2 and SageMaker, with plans to expand cloud support. This indicates they’re running an LLM orchestration layer in cloud environments and iterating publicly Leeroo orchestrator GitHub.

Who are their target customer(s)

• Enterprise data engineering teams at banks/fintechs: They need automation that can run in a VPC/on‑prem without leaking data. Today much of their time goes to wiring dashboards, handling flaky APIs, and fixing data quality incidents instead of automating them Leeroo agents Leeroo company YC page.
• Analytics/BI teams in large enterprises: They have to deliver recurring reports and live dashboards and answer ad‑hoc questions quickly, but lack bandwidth to build robust automation themselves; they want prebuilt analyst/reporting agents they can adapt Leeroo agents.
• ML engineering / model‑ops teams: They struggle to monitor models and agents in production, detect anomalies, and turn noisy alerts into reliable actions. They need integrations with traces/logs and monitoring that improves over time Leeroo orchestrator GitHub Leeroo agents.
• Compliance, security, and data‑governance teams in regulated industries: They need AI that is auditable, versioned, and contained within corporate boundaries. They resist one‑off agent demos and prefer governed, enterprise deployments with clear controls Leeroo company Founding Engineer role.
• Heads of data/analytics at mid‑to‑large portfolio companies: They want reusable automation to scale output without heavy hiring. Many vendor pilots require constant vendor engineering or don’t survive handoff; they want trained agents that become reusable assets across teams YC page Leeroo company.

How would they acquire their first 10, 50, and 100 customers

• First 10: Convert existing pilots in banking, finance, and healthcare into paid, in‑VPC/on‑prem deployments with a tightly scoped engagement and a forward‑deployed engineer; use each success as a detailed case study and reference, capturing exact integration steps and compliance artifacts Leeroo agents Leeroo company.
• First 50: Productize the most common pilot patterns into prebuilt blueprints and a deploy kit (VPC templates, connectors, governance checklists) with a fixed onboarding package; add a small channel of boutique SIs to execute deployments while the core team hardens the learning engine and vertical content.
• First 100: Scale via partners and a lighter self‑serve for less‑regulated teams while maintaining an enterprise track; launch a curated marketplace of audited agent blueprints, invest in automated onboarding/audit artifacts, and build customer success to drive multi‑agent expansions in existing accounts.

What is the rough total addressable market

Top-down context:

Relevant adjacent markets total roughly $51.6B by summing BI software (~$36.6B), enterprise AI assistant/agent software (~$9.8B), MLOps (~$3.0B), and data observability (~$2.14B). This ceiling double‑counts overlap and is best seen as context rather than literal TAM BI AI assistants MLOps Data observability.

Bottom-up calculation:

A conservative, serviceable TAM today is ~$12–20B focused on enterprise data/AI automation. One defensible cut uses: full data observability ($2.14B) + full MLOps ($3.03B) + 10% of BI for automation/orchestration (~$3.66B) + 30% of enterprise AI assistants for data/ops use (~$2.95B), totaling ≈$11.8B; modestly loosening carve‑outs yields ~$15–20B Data observability MLOps BI AI assistants.

Assumptions:

• Use 10% of BI market for automation/orchestration budgets that map to agents, not dashboard licenses.
• Use 30% of enterprise AI assistant spend as data/ops‑relevant; exclude general productivity assistants.
• Treat MLOps and data observability full figures as core (sold to the same teams), recognizing overlap in adjacent totals.

Who are some of their notable competitors

• Monte Carlo: Data observability platform now offering “observability agents” for pipeline and agent telemetry. Overlaps where buyers want automated detection/root‑cause on data quality and need telemetry within their own warehouse/VPC—similar to Leeroo’s data‑quality/monitoring blueprints.
• Arize: Model and agent observability focused on tracing, evaluating, and debugging ML models and production agents. Competes for ML/ops teams turning noisy alerts into actionable diagnostics—an area Leeroo targets with monitoring agents.
• Databricks: Lakehouse Monitoring plus serving and model‑ops inside a unified platform. Large data teams may use Databricks for pipeline and model observability instead of an additional agent layer, overlapping with Leeroo’s enterprise use cases.
• IBM watsonx Orchestrate: Enterprise agent/orchestration with governance, pre‑deployment evaluation, and production monitoring. Targets regulated enterprises that require strict controls and audit—similar buyer profile to Leeroo’s on‑prem/VPC deployments.
• Great Expectations: Open‑source and commercial data‑quality testing and validation. Competes on defining and running automated checks for pipelines; it’s not an agentic, continuously learning teammate but addresses overlapping data‑quality governance needs.