UXin3D Portfolio

Esteban J. Struve T.

Profile photo

AI Tech Lead with 10+ years of experience building XR, AI, and data‑driven systems for manufacturing and healthcare. I specialize in C#, .NET, Python, TypeScript, Azure, computer vision, real‑time interactive software, and increasingly, agentic AI systems.

Today at WeAI, I lead AI development teams helping organizations turn AI adoption into tangible business value — integrating AI agent workflows into existing processes, transforming organizational strategies accordingly, and advising stakeholders at every level on AI agent architecture, knowledge strategy, and the implementation of emerging agentic solutions. I map the most complex challenges across customer cases and infrastructures to compatible AI‑based solutions, tackling GenAI deployments across diverse environments.

I’m passionate about being the AI amplifier: bringing the latest techniques, tools, and practices into the organization to stay ahead of the adoption curve, while empowering every engineer to use AI effectively and consistently produce production‑level software.

My background spans cross‑functional technical leadership, Scrum Master–certified team facilitation, hardware‑agnostic platforms, immersive training and operational tools, and the transformation of complex technical problems into reliable products with measurable impact. I also recruit highly specialized profiles for AI‑assisted coding to make teams grow their capacity within a very uncertain talent market and shifting standards.

Current focus: AI agentic workflows with LangGraph, RAG and evaluation with LlamaIndex, scalable AI engineering systems, and exploring the technologies shaping the next generation of software development.

Want to know something specific? Use the chat window in the lower right corner to ask customized questions about my profile. Just click the circular chat button.

Selected work:

My GitHub

Personal AI Assistant is a private Telegram assistant I use every day: calendar reads and writes, reminders, document Q&A with page-level citations, current web answers, image understanding, and Google Cloud cost reporting. It runs on a stack I can inspect and operate, from the LangGraph agent to the Postgres checkpoints and the billing report.

Python 3.13 · FastAPI · LangGraph · Postgres + pgvector · Google Cloud

Every external write is prepared, never performed automatically. Approval is bound to the exact arguments shown, claimed atomically, and expires; the bot is allow-list only and has a €5/month cost target with alerts at 50/80/100%.

Unlike Muse—whose published paid plans start at $20/month (Power) and $100/month (Maximum)—Meta is not in the data path: Telegram transports messages, Cloud Run and Postgres run in services I choose, and model turns go only to the provider I configure. My €5/month target is metered, reported daily and tuned rather than a subscription price set elsewhere.

Read the case study →

Recreated iPhone Telegram interface: daily conversation, the robot collaborator profile, a cited document answer and an argument-bound approval

Web Agent is a local product research assistant designed to intelligently browse the web and extract structured information about products and pricing. Built with LangGraph and Playwright, it automates e-commerce exploration and leverages the Trustcall library for high-fidelity, structured data extraction. The system features a PostgreSQL-backed long-term memory with semantic search, enabling the agent to recall previous research and perform complex comparisons without redundant browsing. Integrated with LangSmith for execution tracing, it demonstrates a professional agentic workflow for automated market research. Below is the agent's structure running in LangSmith Studio:

LangSmith Studio: agent graph structure, a live execution thread, and the final product result returned to the user

AI Agents 3D Sandbox is a highly experimental Unity project that investigates whether a multimodal LLM can control a character inside a 3D environment — replacing conventional state machines and NavMeshes with language model reasoning. At each iteration the agent observes the scene, sends that context to a model running via Ollama (optionally on a separate PC over the network), receives a high-level decision, executes the resulting movement inside Unity, and carries a small amount of short-term memory into the next cycle. Built with Unity 6 and URP, the sandbox is less about outperforming traditional navigation and more about exploring the capabilities, failure modes, and emergent behaviour of LLM‑driven embodied agents — including how well multimodal models handle spatial reasoning challenges like depth, orientation, and obstacle avoidance.

Unity 3D sandbox: the LLM-controlled agent navigating corridors, approaching obstacles, and reaching different areas of the maze across iterations

Archive:

2026 – Markdown Viewer Archive

A dark-mode Windows markdown viewer and split editor built as a rapid Claude Code prototype, with live rendering, synchronized scrolling and native window behaviour across multiple monitors.

2026 – E‑Commerce Chatbot Archive

A LangGraph ReAct and RAG shopping assistant over a real pet-product catalogue, with a FastAPI backend, grounded recommendations and a responsive product interface. Retired from the selected work because the Telegram assistant now carries the stronger agentic story.

Career timeline:

2025 – Dynamic Holographic UI Generation

Owned the end‑to‑end implementation of a system to improve the platform in production. This AR platform already including advanced cloud features like ML models and hardware integrations like eye‑tracking, but UIs were produced manually. I implemented a dynamic generation system to build holographic XR‑ready UIs that could be authored by code and modified with a fraction of the time, based on customer needs.

AR platform in use at TAP aircraft maintenance, featured on TVI news

2022 – XR Prototyping & EM sensor integration

Built several PoCs with platforms like Meta Quest, Varjo's XR3 and MS HoloLens 2 where an anatomical model could run in the background to simulate patients. They included collaborative simulations built with Photon PUN 2 for multiplayer interactions. Help create a new workflow for XR fast prototyping as Senior Developer and certified Scrum Master, setting the foundations for the product VR Nursing.

XR prototyping session with VR headsets and patient simulation mannequin

2020 – Knowledge Transfer using Mixed Reality experience

Implemented a tool for Manufacturing customers in MS HoloLens (1 & 2), that could record a 3D space and a set of steps with details gathered by voice and camera. The Spatial Anchors were saved in Azure cloud and retrieved later in the same room (by the same device or other) to recover the spatial data and their poses relevant to the process captured. An apprentice could use the app to learn what the expert recorded about the physical space and machinery present in the factory.

Developer wearing HoloLens demonstrating Mixed Reality knowledge transfer application

2019 – Assembly of Satellite Antenna of First AU Nanosatellite mission

Participated as volunteer Engineering student in the Delphini‑1 mission of Aarhus University. Member of the team that assembled the antenna that tracked the position of the nano‑satellite in orbit to send and receive the commands and data payloads from the camera in space. Implemented in Python a friendly UI to track the position of the antenna for the Control Room Software that operated this mission and the following ones.

Delphini-1 satellite mission team with antenna assembly at Aarhus University

2018 – Udacity's VR Developer Nanodegree Final Project

First VR Unity3D project building an immersive experience I called “Spirits of the Jungle”: a survival experience in the Amazon rainforest to showcase my skills as VR developer. It included complex interaction paradigms to navigate the space, speech recognition, climbing up using the controllers and a couple of first‑person shooting scenes to complete the journey. Deployment device was Oculus Rift from PC (predecessor of the Meta Quest headset family).

Spirits of the Jungle VR experience — speech recognition interaction in Amazon rainforest scene

2018 – VR Emergency Room

Along side the Master's Thesis project, I built a PoC for emergency meetings in a corporate setting like banking. The way of handling an emergency repair in the IT infrastructure was to flying in experts from all over to an expensive equipment room to solve the issue at hand in not more than 48 hours. My VR room was a much faster and cheaper solution where these experts already had the hardware distributed and this virtual room could have much better data visualization tools.

VR Emergency Room demo with team of professionals observing a user in a mixed reality headset

2013 – Periodic Pattern Detection and Vortex Synchronization in Turbulent Bidimensional Fluids

Implemented in MATLAB a Computer Vision tool to obtain the vector field of a turbulent fluid considered bidimensional. Coded from scratch particle image velocimetry capable of measuring in the high‑speed video dynamics of a system with chaos, filtering the data in multiple steps to bring up a rotating vector allowing specialized studies such as recurrence and synchronization between the vortexes recorded.

MATLAB computer vision tool output showing vector fields and vortex detection pipeline for turbulent fluid analysis Contact me