Dion David Hands-on CTO
Hands-on CTO, twice: ScoutsCapital now, TrackYourFiles before. Co-founder of ReplyBay. I lead six developers and set the direction, and I still build most of the platform myself, from Kubernetes and Terraform up to the app in your pocket.
Open to advisory and part-time workRemote · Ho Chi Minh City LinkedIn GitHub CV (PDF)
Apps
Edge
Services
AI
Data
Platform
Point at any part of the platform to see what I did there.
How I build now
After ten years of writing every line, I now write very little by hand. Claude Code and Codex do the typing; I own the architecture, review every change and decide what ships. Not vibe coding: written briefs, scoped parallel sessions and gates on every merge. PairBrowse exists because my agents needed a browser I could watch.
Terraform plan for the new node pool
Claude Code in infrastructure
Payments webhook, tests first
Codex in api
Match page performance pass
Claude Code in web app
Second opinion on the API diff
Codex in review
Me: reviewing diffs, deciding what ships
Architecture, priorities, final call
A typical day, simplified.
Every task starts as a written brief and a plan I approve. No prompt-and-pray.
Parallel Claude Code and Codex sessions, each owning one clear piece of work.
Tests, type checks, a second agent reviewing, and my own review before anything merges.
Architecture, trade-offs and the final call stay with me. The agents do the typing.
A problem we caught
Rollouts were getting slower than they should be. With our principal engineer I traced it to two things: the autoscaler reacted to memory instead of CPU, and ArgoCD and the autoscaler were both adjusting replica counts. We moved scaling to CPU, gave the autoscaler sole ownership of replicas and tightened worker limits. Rollouts became fast and predictable again, and about a second of latency per request disappeared.
Selected work
Products in production, with real users and real on-call. Each one has a short write-up of the problem, my role and how it is built.
CTO
A social platform for football with a profile for everyone in the game: players, coaches, scouts, referees, clubs and leagues. Posts, video, live streaming and match results, so talent gets discovered through what it actually does.
Co-founder
AI customer support for online stores. Email, WhatsApp, Instagram, Messenger, TikTok Shop and public comments land in one inbox, and each reply is drafted from the customer's order and the store's own policies.
Creator and lead engineer
One live browser for you, your AI agent and your team. Claude Code or Codex drives the tabs, teammates join with their own cursor, and you take over whenever you want.
All in on ScoutsCapital for three years, with a team that made it real.
Tobias Schaller, Toby, came with a small idea to make football talent more discoverable. When we met, we decided to aim bigger: a social platform built around football, bringing together posts, video, live streaming, match results and profiles for everyone in the game. The kinds of experiences people know from Facebook, TikTok and Instagram, connected to football.
We wanted players to be discovered through what they actually do: their matches, footage and participation, rather than just scores or claims they enter themselves.
Building that with a small team meant finding an architecture that could support our ambition at a fraction of the cost. I focused on making the infrastructure and streaming affordable enough to turn the idea into a working platform.
The early years were 18-hour days. I lived from my savings and took only a small management fee to cover the costs of my own holding company, because I believed in the product and the team. Together we shipped on iOS, Android and web, cut streaming costs by 99% and kept a platform for everyone in football running on a startup budget. Their talent and dedication made that possible.
Our aim is to make football more discoverable and give talent a better chance to be seen.
Dion DavidOn building ScoutsCapital as CTO, since 2023AI and data science
I studied AI for my master's and now ship models that run on every upload. Convolutional networks, recurrent networks and transformers: I go from raw data and features to a trained network, its evaluation, and the infrastructure that serves it.
Models I designed, trained and shipped in PyTorch. Frame-by-frame checks on images and video, and a dual model that reads audio and video together to detect violence. Served from Vertex AI, with a local fallback if the cloud is down and a cache that skips media it has already seen.
Speech-to-text with Whisper and scene detection enrich uploaded video automatically, run as durable Temporal workflows.
Language models review ads, enrich content and translate the app, called straight from the core API.
Evaluated and improved lead scoring and lead value models. Compared the production random forests with a recurrent LSTM network, explained predictions with SHAP and LIME, and fixed heavy class imbalance with SMOTE.
What I do
Small teams need people who can go from a Figma frame to a Kubernetes cluster without a handover. That is the job I like best.
My strongest area. Production platforms that scale and stay up: multi-cluster Kubernetes, everything as code, GitOps deploys and full observability.
In production on Google Cloud: GKE, Cloud SQL, Vertex AI, Pub/Sub, BigQuery and Secret Manager. Terraform and Kubernetes are cloud-agnostic, so the same setup runs on AWS (EKS), Azure or your own hardware.
I design products as well as build them: once in Figma, now straight in code.
From data and features to trained neural networks in production, plus agents that do real work with people in the loop.
Where the technology goes next: architecture, platform choices, roadmap and the trade-offs between speed and stability.
75 tools I have used over ten years. Hover to pause.
Experience
2023 - now
Lead a team of six developers and work alongside the CEO, CRO and CMO and their teams. Decide where the technology goes next: architecture, platform, security and the roadmap, while still building hands-on.
2026 - now
Co-founded an AI customer support platform for online stores. Build the AI reply pipeline, API, web apps and EU infrastructure.
2014 - now
Years of side projects across web, mobile, infrastructure and ML, each new stack learned properly by shipping something real with it, fast.
2023
Built AI features and web applications with cross-functional product teams.
2018 - 2022
Ran technology strategy and led development of the file tracking platform for four years.
Education
2018 - 2021
Joint degree: VU Amsterdam and University of Amsterdam
Amsterdam, Netherlands
Joint master's of VU Amsterdam and the University of Amsterdam, on the Software Engineering and Green IT track. Algorithms, data structures and system architecture, with a final specialisation and thesis in AI.
Master's thesis
Evaluating and Improving Existing Marketing Related AI/ML Models
An industry case study with Exsell, a predictive marketing company. I evaluated their LeadScore and LeadValue models, which find likely new B2B customers, and tested a recurrent neural network (LSTM) against the existing random forests. Feature contributions were explained with SHAP and LIME, and SMOTE handled the heavily imbalanced data.
Supervised by Prof. Dr. Fabio Massacci · defended August 2021
2014 - 2018
Amsterdam University of Applied Sciences (HvA)
Amsterdam, Netherlands
Software engineering fundamentals through hands-on projects in many languages and tools.
2019
Korea University
Seoul, South Korea
Software development, virtualization (KVM), security and machine learning.
2017
Seoul National University
Seoul, South Korea
An intensive program in modern development methods and advanced programming.
Contact
I am staying at ScoutsCapital and ReplyBay, and open to advisory and part-time work: architecture and infrastructure reviews, AI and agent systems, or a hard problem that needs a second pair of hands. I read everything myself.
Pinecone
Chroma
ComfyUI