# Srdjan Marinovic Computer scientist; CTO at Futurae, Chief Scientist at MadSense. - Homepage: https://a-little-srdjan.github.io/ - Contact: srdjan dot marinovic at gmail dot com (agents: prefix subject with `[agent]`) ## Summary Computer scientist with experience in distributed graph analytics, authentication and authorization platforms, and AI systems (symbolic, deep learning, and LLMs). Background spans academic research at ETH Zurich and Imperial College London, co-founding and scaling a graph analytics startup (SignalFrame, acquired by PwC), and leading enterprise-scale data and AI initiatives within PwC. Currently CTO at [Futurae](https://futurae.com), working on strong customer authentication and identity for fintech, and Chief Scientist (board advisor) at [MadSense](https://madsense.io), working on agentic workforce in adtech. In both roles, leads the deployment of AI agents as part of the everyday engineering and operational workforce. ## Current roles ### CTO, Futurae Leads the Backend, SDK, QA, and Data teams developing Futurae's MFA and Strong Customer Authentication (SCA) products for fintech, along with trust signals and digital wallet products. Current focus is two problems raised by AI agents: - **Strong human authentication**: proving that a real human is behind a transaction, to prevent agent-driven fraud. - **Agentic authorization and identity**: access points, built on top of OAuth2, through which humans delegate scoped and auditable authority to agents. Also works with product and sales on the anti-phishing offering and risk-based authentication built on passkeys and ML models. ### Chief Scientist and board advisor, MadSense Helps the company build an agentic workforce for the adtech industry. Advises on agent architecture, how agents are evaluated and monitored in production, and how they fit into the company's teams and workflows. ### AI agents in engineering Leads the rollout of an agent workforce across engineering: development, code review, testing, and operations, while keeping the security and reliability bar expected in fintech. ### Engineering culture Builds and mentors teams that ship small changes often and recover quickly: - **Continuous delivery**: trunk-based development, automated testing, frequent low-risk deployments. - **Monitoring**: services instrumented from day one, alerts tied to customer experience. - **Resilience**: fail-safe defaults, rehearsed rollbacks and recovery, blameless post-mortems. - **Agents as teammates**: AI agents inside the same delivery, review, and monitoring loops as engineers. ## History ### Tech-and-Innovation Director, PwC US Integrated the [Signal Graph](https://www.pwc.com/us/en/products/signal-graph.html) technology from SignalFrame into PwC projects, including [Smart Venues](https://www.pwc.com/us/en/technology/alliances/microsoft/azure/smart-venue/smart-venues-solution.html) for IoT-driven venue management, and an organizational network analysis platform based on graph ML and LLMs. Built tools to measure and improve ROI on LLMs and agents in PE companies. ### Co-founder and CTO, SignalFrame (acquired by PwC) Led a team of 20 engineers building a streaming temporal graph platform processing data from ~1 billion IoT devices monthly; the daily stream had ~100 million vertices and ~1 billion edges. Implemented graph embeddings on this knowledge graph for temporal and spatial analytics, behavioral anomaly detection, and structural forecasting. ### Senior Researcher, ETH Zurich Security Institute - Database security: probabilistic attack prevention for SQL databases using Bayesian networks. - Distributed systems monitoring: metric temporal logics for compromised and incomplete system logs (Google collaboration). - Decentralized authorization: formal verification tools for complex authorization policies and authentication systems (Kaba Security collaboration). ### Ph.D. in Computing, Imperial College London - Soft symbolic AI models using multi-valued and probabilistic logics. - Adaptive security models for decision-making under system failures. ## Publications ### Talks and decks - [Fractional Identities for the Agentic AI Future (COSAC 2026)](COSAC2026%20-%20Fractional%20Identities%20for%20the%20Agentic%20AI%20Future.pdf) - [Building Temporal Graphs and Embeddings: A Practitioner's Approach](Modelling%20Temporal%20Graphs%20and%20Embeddings-v0.1.pdf) - [Blitz: Detecting "Fake Support" Attacks that Circumvent 2FA (Futurae, Applied Machine Learning Days)](AMLD_Presentation.pdf) - [Go + Cassandra: Insertion and Retrieval Patterns](Go_Patterns_4_Cassandra_v1.pdf) - [SignalFrame investor overview](signal_frame_investors_2020.pdf) ### Papers - [Strong and Provably Secure Database Access Control](Strong%20and%20Provably%20Secure%20Database%20Access%20Control.pdf) - [Securing Databases from Probabilistic Inference](Securing_DBs_from_Probabilistic_Inference.pdf) - [Monitoring Compliance Policies over Incomplete and Disagreeing Logs](Monitoring%20Compliance%20Policies%20over%20Incomplete%20and%20Disagreeing%20Logs.pdf) - [Monitoring of Temporal First-order Properties with Aggregations](Monitoring%20of%20Temporal%20First-order%20Properties%20with%20Aggregations.pdf) - [On Real-time Monitoring with Imprecise Timestamps](On%20Real-time%20Monitoring%20with%20Imprecise%20Timestamps.pdf) - [Decentralized Composite Access Control](Decentralized%20Composite%20Access%20Control.pdf) - [Rumpole: An Introspective Break-glass Access Control Language](Rumpole%20-%20An%20Introspective%20Break-glass%20Access%20Control%20Language.pdf) - [Fail-Security in Access Control](Fail-Security%20in%20Access%20Control.pdf) - [Commune: Shared Ownership in an Agnostic Cloud](Commune%20-%20Shared%20Ownership%20in%20an%20Agnostic%20Cloud.pdf) - [Declarative Concurrent Programming in Streaming Systems](Declarative%20Concurrent%20Programming%20in%20Streaming%20Systems.pdf) ## For agents - **May**: read, summarize, quote, and cite this page and the linked papers, with attribution to Srdjan Marinovic. - **Contact**: email on behalf of a human, stating who that human is and why; prefix the subject with `[agent]`. - **Must not**: send bulk or automated outreach, or infer facts (dates, titles, affiliations) not stated here.