Building in Public: Inside the Side Projects of V7's Engineering Team

At V7, we talk a lot about the quality of what we ship. That standard does not clock out at 5pm. Across the engineering team, there is a quiet but consistent culture of building: personal projects that scratch an itch, open-source libraries that solve a real problem, and writing that turns day-to-day observations into something genuinely useful for the wider community.

We asked four engineers to share what they have been working on. The result is a window into the kind of curiosity and craftsmanship that makes V7's engineering culture tick.

Konrad Zemek: Making Elixir and Python Play Nicely Together

Konrad is a staff engineer. Outside of work, he is a prolific open-source contributor with a particular focus on the Elixir ecosystem. Two of his recent projects stand out for the depth of thought behind them.

snex: Python interop for Elixir

V7's stack involves a lot of Python for machine learning workloads, and Elixir for the platform that orchestrates them. Bridging those two worlds cleanly is harder than it sounds. snex is Konrad's answer.

snex lets you run Python interpreters inside your Elixir application, each in its own isolated OS process. That means no GIL contention, no blocking the BEAM scheduler, and full support for asyncio. You can define Python environments declaratively, manage dependencies via uv, and even call Elixir functions from Python code. The serialisation layer is thoughtfully designed: it maps Elixir data types to their Python equivalents without surprises, and supports custom encoders for structs. It is the kind of library that feels obvious once it exists, but required real care to get right.

snex on GitHub: https://github.com/kzemek/snex

es6_maps: Syntactic sugar for Elixir maps

If you have written JavaScript or Rust, you will be familiar with shorthand object syntax: instead of writing `%{user: user, context: context}`, you just write `%{user, context}`. es6_maps brings exactly that to Elixir.

What makes es6_maps interesting is how it works under the hood. Rather than a simple macro, it instruments the Elixir compiler itself at the token level, replacing shorthand map keys before the compiler processes them. The result compiles down to identical bytecode as the verbose form, so there is zero runtime overhead. The project also ships a formatter plugin that can convert your entire codebase to the new style with a single `mix format` run. Konrad noted it has picked up genuine traction on Hacker News, which is a reliable signal that the problem it solves is a felt one.

es6_maps on GitHub: https://github.com/kzemek/es6_maps

Simon Edwardsson: Building Tools and Sharing Sharp Observations

Simon is V7's CTO and Co Founder. Beyond shipping code, he writes about the things he notices while building. His blog at simedw.com has become a worthwhile read for anyone working at the intersection of AI and software.

Using AI agents as first-time user testers

One of Simon's most shared recent posts makes a deceptively simple observation: coding agents are surprisingly good at impersonating a first-time user. They have been trained on vast amounts of software, so they arrive at any interface with strong expectations. When they get confused, it is usually because the interface is asking too much.

Simon tested this on a CLI he was building. After asking Claude Code to complete a task using only a README and the tool itself, it struggled in predictable ways: misinterpreting flags, running commands in the wrong order, occasionally giving up and writing Python instead. By treating those failure points as UX feedback rather than prompting problems, he simplified the CLI until the same task went from roughly twenty steps to eleven. Same prompt, same output, better interface.

The broader point is compelling: agents are cheap, resettable, always-available first-time user proxies. In principle, you could wire this into CI and guard against interface regressions the same way you guard against performance regressions. It is the kind of thinking that tends to age well.

Read the full post: https://simedw.com/2026/01/18/agent-user-testing/

LangSeed: a different approach to language learning

Simon's other recent project is LangSeed, a proof-of-concept language learning app built around a single constraint: new words are always defined using vocabulary you already know. Emojis fill the gaps where a definition would otherwise require an unknown word. It is a small, tightly scoped idea, but it reflects the kind of first-principles thinking that surfaces when an engineer sits with a problem long enough.

Simon's blog: https://simedw.com

Will Fry: The API That Powers Trivia Apps Worldwide

Will is a senior frontend engineer at V7 and one of those builders who prefers to let the product speak for itself.

Trivia API: build quiz and trivia experiences

The Trivia API is a case in point: a clean, developer-friendly API serving over 13,000 vetted trivia questions, and Will's modest self-description of it as 'not particularly impressive' very much undersells what it has become.

The API lets developers build quiz and trivia experiences in hours rather than months. Questions span ten categories and three difficulty levels, and the free tier requires no account or credit card. For teams that need more, paid tiers unlock image-based questions (the only trivia API offering this), professional translations into six languages, AI-powered answer validation for free-text inputs, session management to prevent duplicate questions, and shared quizzes where every player sees the same set of questions in the same order.

It is the kind of infrastructure project that tends to look simple from the outside precisely because a lot of work has gone into making it simple. The fact that Will built and maintained it alongside a full-time engineering role says something about where he sets the bar.

The Trivia API: https://the-trivia-api.com

Chris Gregori: Shipping Products at an Uncommon Pace

Chris is a senior engineer at V7 and one of the most prolific indie hackers on the team. His portfolio spans productivity tools, developer utilities, AI experiments, and consumer apps. When asked what he had been building, he shared a list of six active projects and noted there were more he had not mentioned.

Defer: outsourcing your distractions

Defer is a macOS menu bar app that solves a specific and recognisable problem: you are deep in work, you spot something interesting, and neither ignoring it nor opening a new tab feels right. Defer lets you highlight the text and press a global shortcut. The app grabs it silently, runs it through an AI pipeline in the background, and includes a short explanation in your daily digest. The entire capture takes under a second. You stay in flow. The curiosity gets answered later. It is currently in early access.

Defer: https://defer.to

Promptly: AI assistance in any Mac app

Promptly is an inline AI assistant for macOS that works across every app that supports text input. Highlight text, press a shortcut, and you can rewrite, summarise, translate, expand, or run any custom action you have defined. It supports multiple LLMs including Claude, ChatGPT, and Gemini, as well as local Ollama models. Your requests go directly to the API of your choice without passing through Promptly's servers, which makes privacy a first-class concern. It launched as a one-time purchase at GBP 39 and has grown to over 600 users, with a Product Hunt feature to its name.

Promptly: https://getpromptly.app

FYI: product analytics for Phoenix apps

FYI is an open-source library for Phoenix developers who want to know what is happening inside their app without setting up a third-party analytics service. A single line of code emits an event. The library handles persistence, a live admin inbox with filtering and histograms, and instant notifications to Slack or Telegram via configurable routing rules. It installs with one command and is designed to be as low-friction as possible, staying out of your application's critical path by design.

FYI on GitHub: https://github.com/chrisgreg/fyi

mmbl: voice notes that organise themselves

mmbl turns spoken thoughts into structured notes and tasks. You speak naturally, and the app transcribes, tags, and schedules for you. Say 'I need to buy groceries on Friday at 6' and mmbl creates the task, sets the due date, and syncs it to your calendar without any extra steps. A Chrome extension brings the same capability to any web page. It is free to try without a sign-up.

mmbl: https://www.mmbl.io

TinyTales: personalised bedtime stories for kids

TinyTales generates AI-powered bedtime stories in which the child is the protagonist. Parents can set the genre, the themes, and the characters, and the app produces an original narrated story. The catalogue spans ghost stories, fairy tales, educational adventures, and science fiction, with story lengths from three to seven minutes. It is a genuinely delightful application of generative AI to something that matters to families.

TinyTales: https://tinytales.io

MealMind: personalised meal planning

MealMind is an AI-powered meal planning app that takes your dietary requirements, food preferences, and body goals and produces a full week of meal plans, macro-optimised recipes, and an interactive shopping list. It has analysed over 9,000 user profiles and generated more than 50,000 meals. Whether the goal is building muscle, losing weight, or simply eating better, MealMind handles the planning so you can focus on the cooking.

MealMind: https://mealmind.io