Enterprise generative AI
My Esko AI
A customer-facing AI solution with grounded retrieval, specialist orchestration and escalation when an issue remains unresolved.
I build AI systems that help people find answers in enterprise data and make sense of complex information.
At Esko, I lead applied AI engineering. My earlier research in behavioural biometrics shaped how I approach modelling, experimentation and evaluation.

Selected work
Three featured projects, with six more spanning prototypes, research and educational software.
Enterprise generative AI
A customer-facing AI solution with grounded retrieval, specialist orchestration and escalation when an issue remains unresolved.
Behavioural biometrics
Research on recognising mobile users through typing behaviour, with biometric modelling, benchmark evaluation and peer-reviewed publications.
Collaborative research
International EU research on privacy and safety in the social web, spanning NLP, behavioural analysis and privacy-aware interaction.
Research
Publications on mobile biometrics, image processing and security, with source links and citations.
2021
IET Biometrics, 10(1)
2020
43rd International Conference on Telecommunications and Signal Processing (TSP)
2019
8th European Workshop on Visual Information Processing (EUVIP)
2015
National Conference on Computing, Communication and Information Processing (NCCCIP)
2014
International Conference on Issues and Challenges in Intelligent Computing Techniques (ICICT)
Expertise
Four areas of practice, from preparing data to building and evaluating AI applications.
Design retrieval over enterprise knowledge using hybrid search and agentic RAG, with Model Context Protocol (MCP) integrations that connect models to tools and data.
Extract and organise information from documents, artwork and images using NLP, multilingual processing and computer vision.
Prepare data and build pipelines in Databricks, enrich metadata and knowledge graphs, and present findings through Power BI dashboards and analytical reports.
Build secure FastAPI services and React interfaces, then evaluate response quality using test datasets, telemetry and user feedback.
My work on My Esko AI spans retrieval design and agent orchestration through to FastAPI services and the React interface. I work with Azure OpenAI and Azure AI Search to connect product knowledge to customer questions, with conversation state, source references and feedback captured as part of the experience.
I contribute to evaluation and monitoring alongside implementation, using test datasets and user feedback to guide revisions. Earlier at Ascendum (opens in a new tab), I developed prototypes for medical-report classification and summarisation, recommendation and other applied-AI use cases, taking them into solution proposals and client demonstrations.
At Roma Tre University (opens in a new tab), I built and evaluated models for recognising mobile users from touch interactions (keystrokes, swipes and related gestures), including Gaussian mixture models and convolutional neural networks. This involved experiments on mobile touch datasets, comparison of biometric performance and peer-reviewed publications, including Deepkey and a review within the broader field of mobile touch research.
Through ENCASE (opens in a new tab), I worked on ML/NLP modules, EEG analysis and sensitive-content filtering within privacy and online-safety research. A later UI/UX internship at TimeVillage involved personas, empathy maps and user journeys, adding a practical way to examine how people move through an interface.
Milestones
A timeline of engineering roles, research collaborations, publications and education.
Promoted to lead applied-AI architecture and engineering at Esko (opens in a new tab), supporting secure enterprise adoption across Veralto (opens in a new tab).
Joined Esko (opens in a new tab) within Veralto (opens in a new tab) to build and deliver production AI systems.
At Ascendum (opens in a new tab), applied generative AI, computer vision, recommendation, and predictive analytics across healthcare, e-commerce, and EV batteries.
First journal acceptance in IET Biometrics consolidated research on mobile behavioural biometrics into a structured review of the field.
UI/UX research at TimeVillage focused on personas, user journeys and interaction flows.
An ENCASE (opens in a new tab) secondment at Signal Generix (opens in a new tab) brought behavioural-security research into an applied international R&D environment.
An ENCASE (opens in a new tab) secondment at Telefonica I+D (opens in a new tab) provided exposure to large-scale industrial research in privacy, security, and deployment.
Helping organise the 26th EUSIPCO (opens in a new tab) involved conference organisation and support for the international research community.
Studied contemporary biometric methods at the 15th IAPR/EURASIP International Summer School.
An ENCASE (opens in a new tab) secondment at CyRIC (opens in a new tab) added experience translating security research within an innovation-driven applied technology environment.
GPU programming training at Cineca (opens in a new tab) covered parallel computing for deep-learning and CNN workloads.
At UNIROMA3 (opens in a new tab), built and evaluated mobile behavioural-biometric methods using keystroke dynamics, GMMs, and CNNs through ENCASE (opens in a new tab).
At the IIT Guwahati (opens in a new tab) Virtual Labs initiative, developed remote lab experiences, explored an anthropology/HCI lab, and guided four undergraduate interns.
Teaching at Kendriya Vidyalaya (opens in a new tab) built an early strength in explaining technical concepts clearly to different learners.
At NERIST (opens in a new tab), completed a thesis on reversible secret image sharing using DHWT in MATLAB.
The first IEEE conference paper marked an early transition from classroom engineering into peer-reviewed technical research communication.
The B.Tech at ADBU (opens in a new tab) covered programming, systems and software engineering.
Interactive studio
Explore retrieval, orchestration and grounded responses in the Agentic RAG Studio.
Choose a mission to follow intake, routing, context planning, specialist work and response synthesis. This is an interactive simulation; answers and confidence scores are illustrative, with no live AI service connected.
A little context. A smarter answer.
CHOOSE A MISSION
Each card launches a controlled AI workflow scenario.
Tap a mission to explore its workflow.
From context to a useful answer.
Choose a scenario and follow each step as the agents work together.
Choose a scenario to manually launch an agentic workflow.
Load the selected scenario and identify what the user is trying to achieve.
Classify the domain, success criteria, risk level, and routing confidence.
Prepare the sources, signals, checks, and output format before delegation.
Select a scenario to reveal the delegated specialist and workflow context.
Generate the final answer gradually with traceable reasoning and next actions.
Your mission, brought together.
Run a mission to see the specialist’s response and suggested next steps.
Your agents are working on this mission.
No agent is selected by default. Choose one of the scenarios to start the workflow manually.
Photography
People, places and moments beyond the screen.
Photograph 1 of 10: Windowed Silence
Contact
For AI products, research collaborations and engineering challenges. I’m preparing Medium articles on translating LLM optimisation research into working implementations. The articles are currently in development.
This portfolio brings together colour theory, Figma design, responsive web development, Git CICD, and Azure deployment. If you’re exploring a similar project, I’m open to sharing guidance and collaborating.
Never stop fighting until you arrive at your destined place — that is, the unique you. Have an aim in life, continuously acquire knowledge, work hard, and have perseverance to realize the great life.