Leonardo Martins
Verified Expert in Engineering
Principal Consultant and Developer
London, United Kingdom
Toptal member since May 19, 2026
Leonardo is a principal consultant and "archiveloper" with 20+ years of experience at Barclays, HSBC, Credit Suisse, and HMPO. He combines enterprise architecture with hands-on engineering to design and deliver AI-native research platforms in regulated environments. A London Business School Executive MBA holder and inventor of patented AI optimization technology (GB2541570), Leonardo specializes in agentic AI, MCP, AWS Bedrock, SageMaker, RAG, Databricks, and cloud-native data platforms.
Portfolio
Experience
- Linux - 20 years
- Docker - 10 years
- AWS IoT - 10 years
- C - 10 years
- Artificial Intelligence (AI) - 10 years
- AWS Lambda - 7 years
- Python 3 - 5 years
- Delta Lake - 3 years
Preferred Environment
Windows, Linux, Visual Studio Code (VS Code), Windows Subsystem for Linux (WSL), Docker, AWS IoT, Azure, Python 3, Java, TypeScript 5
The most amazing...
...thing I've achieved was being granted a UK patent GB2541570 in 2015 for an AI system that automatically tunes high-frequency trading applications.
Work Experience
Lead Architect
HM Passport Office
- Oversaw architecture and delivery for critical national infrastructure at HMPO via Pontus Vision as the lead architect for BI, MI, and AI platforms.
- Reported to the CTO, maintained security clearance (SC), and oversaw operations for one of the UK's largest public-facing government services.
- Stabilized a failing BI/MI infrastructure that was experiencing daily outages, restoring service reliability and establishing HMPO as one of the most reliable passport services in the UK government post-pandemic.
- Designed and delivered a unified data layer using Databricks, aggregating data from every step of the passport application journey, enabling accurate SLA measurement and bottleneck identification for the first time.
- Introduced serverless architecture on AWS, migrating critical national infrastructure from on-premise to the cloud.
- Pioneered AI adoption at HMPO. Built MCP server wrappers integrating AWS Glue, Databricks, and SageMaker-hosted LLMs, enabling AI agents to programmatically query workforce planning and performance systems.
- Delivered a fixed-cost workforce planning and performance solution that directly improved passport processing times from weeks to days.
Founder | MD | CTO
Pontus Vision
- Delivered high-stakes AI and cloud architecture engagements for financial services and regulated public sector clients.
- Oversaw architecture for the BI/MI and AI platform at HMPO.
- Designed a strategy aggregator for financial market analysis at Private Wealth Management, utilizing NotebookLM, Qdrant, and Amazon Bedrock. Transformed fragmented market data into structured insights for quantitative analysis.
- Built early MCP services enabling AI agents to programmatically query authoritative financial data sources and share research outputs across teams.
- Architected HMPO's first fully serverless AWS repository on critical national infrastructure, securing the 2022 Computing Cloud Excellence Award.
- Designed and built a real-time audio analysis bot using custom ML pipelines to distinguish human from AI-generated voices.
- Secured the Fintech Innovation Lab (2015) award by Accenture.
- Implemented Patent GB2541570 concerning thread performance optimization, specifically focusing on AI-driven optimization of software execution threads.
- Utilized a cascade forward artificial neural network system to enhance thread allocation in high-frequency trading applications.
Low-latency Senior Architect
HSBC UK
- Served as a lead architect for HSBC's first eFX co-location project, building the data connectivity foundation for high-frequency pricing engines across London, New York, and Tokyo simultaneously.
- Reduced latency spikes by 25x, directly improving the research-to-production path for pricing models across FX and equities.
- Built sub-millisecond data connectivity infrastructure, ensuring researchers had real-time access to complex cross-asset market datasets.
- Optimized the end-to-end trading infrastructure across 3 of the world's largest financial centers, including London, New York, and Tokyo, simultaneously.
- Established the performance baseline that enabled HSBC's eFX business to compete effectively in co-location environments against Tier 1 investment banks.
Senior Consultant, Fixed Income Trading Infrastructure
Barclays Capital
- Led a recovery program that saved Barclays Capital over £10 million by eliminating trading floor outages across global front-office platforms.
- Directed a global middleware upgrade program across 30+ engineers, 40 applications, and 7 business units, including fixed income, FX, commodities, STP, BARX eCommerce, BarCap Trader, and secondary credit trading.
- Introduced the bank's first Linux-based application, reducing hardware capex by 80% and increasing platform performance by 400%.
- Architected and delivered the next-generation middleware platform for all trading systems, including fixed income, FX, commodities, and equities, across London, New York, and Tokyo simultaneously.
- Managed a 300% growth in infrastructure capacity over 3 years without service disruption.
- Increased resource fungibility across all business units by leading a cross-training program for over 100 employees.
- Led contingency and capacity management for fixed income's market data distribution infrastructure, reporting to the director of fixed income IT.
Experience
AI vs Human Voice Detector for Microsoft Teams
The bot joins Teams calls as a meeting participant, captures the audio stream, and routes it asynchronously through a Python-based ML pipeline. The classifier is an XGBoost model trained on engineered audio features—mel-frequency cepstral coefficients (MFCCs), spectral statistics, and prosodic features, including pitch, jitter, and shimmer that reveal characteristic artifacts in current synthetic-voice models. The deployed model achieves a precision of 0.85 and a recall of 0.85 for the AI-generated class, balancing false positives against missed detections in a domain where both errors carry real operational costs.
Architecturally, the design is built to escalate from supervised classification today to unsupervised anomaly detection as adversaries adapt—today's labeled training set of AI-generated voices becomes tomorrow's incomplete reference as deepfake technology improves. The whole pipeline sits inside a regulated environment with audit trails, identity controls, and assurance requirements characteristic of UK central government delivery.
Education
Master of Business Administration (MBA) in Finance and Entrepreneurship
London Business School - London, England
Bachelor's Degree in Computer Science and Computer Engineering
Colorado Technical University - Colorado, USA
Certifications
Machine Learning
Coursera
Skills
Libraries/APIs
XGBoost
Tools
AWS Step Functions, Terraform, TIBCO Rendezvous, AWS IAM, AWS Glue, TIBCO Enterprise Message Service (EMS), Amazon OpenSearch, AWS Cloud Development Kit (CDK), Amazon SageMaker
Languages
C, Python 3, Java, TypeScript 5, C++, Assembly, C#.NET
Platforms
Linux, AWS IoT, AWS Lambda, Windows, Visual Studio Code (VS Code), Docker, Azure, Databricks, Azure Functions, Kubernetes
Paradigms
Model Context Protocol (MCP)
Storage
Databases
Other
Artificial Intelligence (AI), Low Latency, SmartSockets, Large Language Models (LLMs), Machine Learning, RAG Systems, Vector Databases, Windows Subsystem for Linux (WSL), Business Plan Consulting, Project Portfolio Management (PPM), Amazon RDS, Amazon API Gateway, FX pricing engines, Solace, Azure Databricks, Amazon Kinesis, Delta Lake, Networking, AWS ECS Fargate, Hugging Face, Colocation, Amazon Bedrock AgentCore, Qdrant
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring