Elliptic delivers blockchain analytics and digital asset compliance at a global scale, helping financial institutions, crypto businesses and government agencies detect, trace and investigate illicit activity across digital asset transactions.
Building on its existing Data & Intelligence Platform, Elliptic leveraged Agent Bricks to build Elliptic’s copilot, designed to reduce compliance analyst workload by describing wallet activity, highlighting exposure paths and producing neutral, report-ready risk narratives, all aligned to client risk appetites. By building on Databricks, Elliptic avoided the significant effort of developing governance and tracing infrastructure internally, accelerating the path from experimentation to a compliant LLM response. With Agent Bricks, Elliptic delivers trusted, regulator-ready insights from complex blockchain data, building a foundation for faster, more consistent AI-assisted dispositions.
Scaling compliance without compromising trust
Rapid growth across crypto markets, rising transaction volumes and expanding regulation place additional pressure on Elliptic’s clients: compliance teams at exchanges, banks, payment providers and institutional custody platforms. Even with robust risk scores and comprehensive screening results, compliance analysts must document their decisions by bringing together information on exposure values, crypto assets, addresses and the chronology of transactions. This makes it harder to scale rapidly without increasing costs.
Elliptic helps compliance analysts spend less time triaging alerts and more time making consistent decisions as crypto adoption drives higher alert volumes, without lowering the bar for regulatory readiness. “Clients need clear, report‑ready narratives that can stand up to regulatory scrutiny,” said Amar Chandarana, Senior Product Manager at Elliptic. “Elliptic’s copilot analyzes and summarizes wallet and transaction activity, highlighting critical exposure and providing relevant risk context for faster decision-making.”
Applying large language models (LLMs) to these workflows introduces new challenges, especially in sensitive domains like sanctions, terrorist financing and child exploitation. Compliance teams needed strong assurances that Elliptic’s copilot would not hallucinate, omit key context, or rely on opaque reasoning, particularly when its outputs might feed into formal investigations and regulatory reports.
To address this, Elliptic used Databricks tooling to make every response traceable, evaluated and governed. MLflow tracing captures prompts, intermediate steps and final narratives for each interaction, so that if an issue is reported, Elliptic's team can quickly understand how a response was created and what information was provided to the model.
Elliptic has defined domain-specific safety guidelines to distinguish between describing financial crime typologies and inadvertently mentioning harmful activity, ensuring Elliptic's copilot explains patterns related to money laundering or sanctions in neutral, factual language tied to on-chain evidence and internal risk models. This combination of traceability, automated evaluation and domain-tuned safety controls allows Elliptic to deploy LLMs in high-stakes compliance workflows with confidence that outputs are consistent, defensible and suitable for regulator-facing use cases.
Extending the Databricks Platform with Agent Bricks
As the team moved from experimentation to a compliance-critical application, Elliptic adopted Agent Bricks to operationalize governance, evaluation and observability without disrupting existing architecture. Agent Bricks enabled the team to embed AI into production workflows while maintaining the rigor required for regulated use cases.
MLflow for traceability and evaluation
MLflow is central to development and oversight. Elliptic’s copilot uses MLflow tracing to capture prompts, responses and metadata for every execution, establishing clear lineage from input to output. Automated evaluation pipelines score responses for correctness, relevance and safety using LLMs as judges. Custom safety scorers were developed to distinguish between explaining money laundering patterns (appropriate) and promoting harmful activities (inappropriate). This allows the team to assess quality systematically and build trust without relying on exhaustive manual reviews.
Madhura Chaganty, Engineering Manager at Elliptic, explained, "Being able to trace and evaluate responses consistently was critical for us. That visibility helped us move faster while holding Elliptic’s copilot to a high standard, as it made it easier to learn from each iteration and make data‑driven improvements."
MLflow also brought structure to prompt and safety management through versioning, auditability and custom safety controls tailored to financial crime contexts. The team leveraged the Databricks Foundational Models API and Model Serving to utilize the full breadth of the Agent Bricks offering, enabling better responses and future-proofing for multi-agent systems. Close collaboration with Databricks informed architectural decisions, including internal testing and client beta feedback loops to refine outputs. Elliptic created a flexible path to improve response quality and prepare for future multi-agent systems without reworking its infrastructure.
From AI skepticism to fast and confident decision-making
By bringing Elliptic’s copilot to market on Databricks, Elliptic delivers immediate value without exhausting resources to build out internal frameworks. Now, analysts can quickly understand sources of risk, exposure details and risk materiality relative to their organization’s risk appetite. Armed with clear, neutral summaries, Elliptic customers can support faster decision-making while maintaining the standards required for internal review and regulatory reporting.
Customer feedback shows meaningful gains in productivity and efficiency. Compliance teams are resolving high-risk screenings up to 2x faster, without sacrificing accuracy or documentation quality. Analysts praise Elliptic’s copilot for clarity of language, balanced tone and the surfacing of data points that might otherwise be missed manually, which has improved analysts’ effectiveness while reducing compliance risk.
Amar explained, “What makes the difference is trust. Clients can see that outputs are grounded in our data, aligned to their risk appetite and built to enhance their existing compliance workflows.”
Beyond the immediate productivity and governance improvements, Elliptic cemented a solid but flexible foundation for developing broader agentic workflows. The team continues to explore new ways to enhance compliance workflows and improve risk detection, with a focus on responsible development in partnership with leading crypto natives. The vision is to trust AI as a productivity enabler, not a risk. Looking ahead, Elliptic plans to continue scaling compliance operations on the Data & Intelligence Platform to satisfy evolving crypto markets without increasing their operational complexity.