Dev Retrospective: A journey of creation, PoCs, and real limitations.
This year was marked by an intense creative journey. Between ideas that served as proofs of concept (PoCs), projects in active development, and some tools that ended up being used by other people, the main result was technical learning.
It all started with the launch of TikTok Live Download, a project that began at the end of last year and was published at the beginning of this year. From there, I delved into various fronts over the months. Most of the projects listed below are unpublished PoCs, created as technical experiments, validation of ideas, or exploration of architecture and protocols.
Projects developed throughout the year
- Pixelfed Brazil
- Remote Control
- Private Social Network
- Auto TV
- Light / Light MQTT
- Hotspot Auth
- Secret NFT
- WhatsApp Clone (Matrix)
- Worktime
- Track Links
- Donate
- Sgame
- Felipe Mateus App V3
- Gmerge
Pixelfed Brazil
A significant portion of my time was dedicated to Pixelfed Brazil, This project, where I performed maintenance, created modules, and did translations into Portuguese, differs from the others because it wasn't something I created from scratch; it's an open-source project, and I made some modifications.
Over time, I chose to stop updating. The relationship with the project's original maintainers isn't the best, and there's always the risk that future changes could block or render my instance unusable. This experience reminded me a lot of... WordPress ecosystemCentralized decision-making and cultural and linguistic barriers significantly hinder collaboration and communication within these communities.
Fedimap (PoC)
O Fedimap It was an experiment inspired by a project I saw on Bluesky: a visual map of the entire Fediverse network, classifying connections between instances. The Proof of Concept (PoC) actually worked, and I was able to generate a functional map.
I abandoned the project because the amount of data is growing rapidly, requiring more and more resources to serve a relatively small bubble. Furthermore, I realized that many instance administrators dislike services that categorize or analyze network data in this way.
Fedipay
O Fedipay It emerged as an attempt to expand the use of the ActivityPub protocol to enable payments within the FeDiverse. The idea was to mediate financial transactions between instances.
During my research, I encountered technical and, mainly, regulatory barriers that make this type of solution unfeasible for an independent developer. A possible use case would be in federated delivery applications: a payment instance regulating the financial flow between restaurant, delivery driver, and customer instances. In addition to processing payments, Fedipay would also define a communication protocol between these instances.
Because it was extremely disruptive and involved money, I ended up abandoning the project.
Opencar (PoC)
O Opencar It was a Proof of Concept (PoC) that started in a very embryonic stage, around 2% to 3% of the project. It was conceived in three parts: passenger app, driver app, and backend. The initial idea was to be completely open source or, at least, to keep the backend open.
The goal was to apply ActivityPub outside the context of social networks, creating a cooperative mobility platform. Each backend would function as an autonomous agent, controlling its passengers, drivers, and partnerships with other agents.
Here's how it works: the passenger requests a ride from their independent agent, who checks if there's service available in the area. If not, the request is forwarded to partner agents. When a driver accepts, the confirmation travels back to the passenger. From there, direct communication between passenger and driver is done via WebRTC, allowing real-time location tracking without the latency of ActivityPub.
In this model, autonomous agents are the actors in the system. The driver is represented as an actor who reports status and approximate location. The passenger is not an actor. The idea was to apply this to transportation cooperatives. The project did not move forward due to lack of time and manpower.
FediGroups
O FediGroups It was an attempt to create support for groups in the Fediverse. The proposal was to define an ActivityPub protocol to transmit groups and related posts, making Pixelfed groups compatible with the rest of the Fediverse.
Remote Control
O Remote Control It's an old project, an evolution of the idea of Sonoff remote control. The goal was to control LG TVs by simulating the official remote control app, operating directly on the local network.
Despite some progress, I still haven't been able to complete the handshake necessary to send commands to the TV. The project remains incomplete.
Private Social Network (PoC)
This project already has a name (not disclosed here) and consists of creating a social network in the style of Instagram, where each user can have their own domain. It has no relation to the Fediverse.
It's a commercial project, currently around 10% completed, and so I won't go into too much detail at this time.
Auto TV (PoC)
O Auto TV Its objective was to automatically generate a video-based TV channel, using a new... IPTV module. The project didn't move forward due to problems like a black screen during execution. Perhaps I'll revisit it in the future with a different approach or using off-the-shelf software for the more complex parts.
Light / Light MQTT (PoC)
This was an experimental project using Kotlin and Go. It captures audio input from the computer, decomposes the sound into binary signals, and classifies them into low, medium, and high frequencies, sending this data via MQTT.
On the phone, the app consumes this data and flashes the flashlight according to the sound signals. The main problems were incorrect conversion of audio to binary, inconsistent values sent to MQTT, and limitations of the Android flashlight API, which does not allow very fast activations or fade effects.
Hotspot Auth
O Hotspot Auth This was a research project to create a product aimed at merchants who want to sell Wi-Fi access. The idea involved using an Android TV Box to control the distribution of Wi-Fi and display advertisements, passing the revenue on to the establishment.
The project was abandoned because recent versions of Android severely limit Wi-Fi control APIs. Currently, this type of control requires the device to be configured as an administrator, which greatly hinders adoption. I also considered implementing this on GPL routers, but I didn't have the necessary hardware available.
Secret NFT
O Secret NFT It's an old idea that has been reformulated over time. Initially, it was intended to be a social network entirely based on NFTs and completely anonymous, but this ran into problems with marketplace moderation and the risk of contracts being banned.
The current version, created this year, is a decentralized web application without a traditional backend, where profiles and posts (text, image, and video) are consumed via IPFS and blockchain. Users can like, share, and possibly comment. Each interaction generates a percentage reward, also benefiting the original author.
To avoid problems with illicit content, all data input goes through a centralized moderation platform before being definitively published on the blockchain. The money raised is used for platform maintenance and profile remuneration. All content is NFT, except for comments and likes. The contract also allows address adjustments in cases of loss of access or fraud, with judicious use of this function.
The project name no longer reflects the original proposal, remaining only as a development codename.
WhatsApp Clone (Matrix)
This project was born as an alternative to WhatsApp, which underwent several changes and forced integrations with other Meta services. The solution uses the Matrix protocol, with Dendrite on the backend.
The frontend is built in Flutter, while the Matrix communication layer was implemented in Rust, since I couldn't find any ready-made libraries that met my needs. This architecture makes development slower, as each new feature requires implementation in Rust, compilation, and testing in Flutter.
Currently, the app already has user registration, profile updates, and message sending capabilities. I'm working on message retrieval and reception. All messages are saved locally on the device, focusing on offline use and future group implementation.
Worktime
O Worktime This is a plugin for VS Code created to solve my difficulty in tracking hours worked per project. The idea was to automatically record the active time on each project open in the editor.
The implementation didn't work as expected, and curiously, other plugins with similar purposes also didn't perform well in my daily use.
Track Links
O Track Links It started as a QR code tracker and evolved into a general link tracking system. It generates tracking URLs, collects IP addresses and user agents, and counts accesses.
Donate
O Donate It was created with the goal of raising funds for the Pixelfed Brazil and the CMS IPTV. It is approximately 90% complete and would allow the creation of projects in the style of “Support”...with one-off donations or subscriptions. It would go live in donate.felipemateus.com, replacing the old donations page.
The system allows for project updates, displays a list of supporters (each supporter can hide their name from the list), and enables the creation of project bundles, allowing a supporter to contribute to more than one project simultaneously. Projects can be donation-based or subscription-based, with recurring payments, and each has a defined goal.
Subscription-type projects have specific settings, and bundles can only be formed with projects of the same type. Projects can only be displayed to authenticated users or only to users with active subscriptions. Donations are made exclusively by credit card, and plan settings are only displayed for subscription-type projects.
I decided not to launch it because I didn't see any real demand. Most people who contact me ask for support or resources, but few show interest in contributing financially.
Sgame
O Sgame (Streaming Game) is a recent Proof of Concept (PoC) that uses FFmpeg to create a live YouTube stream with an interactive chat-based game. Users send numbers in the chat, and at each time interval, the most frequently sent number is chosen. The game indicates whether the correct number is higher or lower, running continuously.
The goal is to increase engagement in live streams on YouTube.
Felipe Mateus App V3
O Felipe Mateus App V3 It's the evolution of an old microblog, as mentioned in the previous post. The new approach transforms the application into an interactive platform. I'm planning to implement an internal currency system, used for interaction and engagement in sub-applications. These currencies are not convertible into money and function exclusively within the app's ecosystem.
Bomb
This will be a sub-app of Felipe Mateus's app. It will function like a minefield where each game lasts an entire month, from the 1st to the 30th. Each square in the game will have bombs with varying purposes, such as being banned from playing for about three days, being banned from playing for an entire month, among other penalties.
In addition, there will be rewards, such as earning coins and promotions. Most of the squares will have no function whatsoever.
The game will work as follows: the player will only be able to make one move per day, respecting the business rules. The database and API for this game will be in a separate system, functioning as a microservice independent of the main app.
Lowest Price
O Lowest Price This is a Proof of Concept (PoC) auction based on the lowest unique bid concept. Bids are only revealed at the end of the auction, and during the process the user only knows if their bid is among the top 10 or not. The goal is to maximize engagement. This is the only sub-app that offers cash prizes.
Blue Corner
The Little Blue Corner It's a live conversation space inspired by Twitter Spaces, but with its own unique dynamic. In it, the host has complete control of the stage, being able to invite and enable only the people they want to speak, while the rest of the audience follows along in real time.
The experience works as a live and interactive podcastThe audience listens, participates via chat, and interacts without directly interfering with the audio.
I developed a proof of concept (PoC) to validate the idea and functionality of the platform. With the tests completed, the next step is to implement the... Blue Corner definitively within the application of Felipe Mateus, integrating this new audio experience into the app's ecosystem.
Gmerge
O Gmerge This was the last project started this year. It modifies the logic of the 2048 game, transforming it into an infinite color matching game. It has a swipe mode, similar to the traditional 2048, and a gravity mode, closer to Tetris. I'm still fixing some bugs, but the game is already available on [platform name]. Google Play.
Conclusion
Ultimately, this wasn't a year about finished products, but about exploration, limitations, and real learning. Many of these projects didn't reach the public, and that doesn't mean failure: each PoC fulfilled its role by revealing technical bottlenecks, regulatory barriers, or simply paths that weren't worth following. Creating, abandoning, refactoring, and starting over were all part of the process. The outcome isn't just in what was published, but in the deep understanding of protocols, distributed architectures, and the ability to make conscious decisions about where not to invest more time. The next cycle begins with fewer illusions and more clarity—and that discernment, in itself, is my greatest achievement.
