Evidence beats narrative.
The site should make a simple promise: Serendeep checks claims by building, reading, measuring, and explaining what he found.
Model-merge forensics and AI essaysAI systems · local-first software · exact technical writing
I build small, useful software and write about the parts of AI and programming that get hand-waved too quickly: model evidence, local tools, mobile architecture, time bugs, and the gap between a claim and the thing that proves it.
Portfolio positioning
The homepage should make one argument clearly: this is a person who cares about proof, working tools, and writing that does not sand off the interesting parts.
The site should make a simple promise: Serendeep checks claims by building, reading, measuring, and explaining what he found.
Model-merge forensics and AI essaysThe projects point toward audio tooling, mobile foundations, time correctness, parsers, automation, and local computation.
GitHub repos and READMEsTwenty RSS-visible posts from May 2025 to July 2026 cover ML, AI policy, local-first apps, Flutter, Linux, and web correctness.
blog.serendeep.tech RSSSpeedcubing and MUN add discipline, debate, and public speaking. They should appear as texture, not as a hobby collage.
WCA profile and Open Minds MUN postFeatured modules
Each item has a source URL and a concrete reason to exist. No invented impact metrics, no shiny icons, no vague claims about changing the world.
A piece on model-merge forensics: using algebra and weights to tell whether an AI model is original, merged, fine-tuned, or wrapped.
LLM evidenceTool / 2026Downloads and analyzes Spotify playlists locally, then sorts tracks into DJ-ready crates by genre, tempo, BPM, and key.
Audio MLTemplate / 2026A production-ready app foundation: BLoC, flavors, FCM, themes, routing, logging, error handling, and environment config.
Mobile infraApp / 2026A React and Temporal API scheduling app for creating events in one timezone and viewing them correctly in another.
CorrectnessRust / 2026A Rust library and CLI for parsing, formatting, manipulating, and converting Valve's KV3 data format to JSON.
SystemsVisual system
The graphics should feel like instruments, traces, plots, and calibration marks. Not blobs, stock gradients, mascots, emoji, or fake dashboards.
Implementation plan
The first version does not need a server. Next gives clean routing, metadata, React components, and an easy path to MDX or RSS-backed writing pages later.
Hero, positioning spine, featured work, latest writing, and contact.
A future RSS or MDX backed index grouped by ML, local-first, Flutter, Linux, and web correctness.
GitHub backed project records with short hand-written notes and source links.
Source-backed content
Computer science student, technical consultant, programmer, writer, speedcuber, and occasional MUN chair. The work sits somewhere between AI systems, app foundations, local-first tools, and careful technical explanation.