Apps & APIs
Scheduling, auth, Discord bots, Docker. ttgod-scheduler is the project where I figured out how to build something other people could actually run.
Samiul Hossain
I like building things that feel useful: full-stack apps, long-running side projects, and lately experiments running open models on my own machine.
01
About
I'm a student who learns by shipping. Most of what I know came from real projects, coursework, and the kind of debugging sessions that run late into the evening.
These days I'm split between product work (APIs, TypeScript, deployment) and seeing how far I can push models on my own hardware without reaching for a cloud API first.
Games & systems
Cinderfall has been a long Godot build: strategy, tactics, and the patience that comes with keeping a messy codebase alive until it makes sense.
Local ML
Running models locally, comparing quantizations, and being honest about when on-device inference is worth it versus when it is still a stretch.
02
Focus
If something can run on my laptop, I'd rather figure that out than sign up for another API key.
Models on your machine, not only in the cloud.
- Privacy matters. Useful software shouldn't need to send everything to someone else's server.
- Open weights and good quantization already make strong models runnable on normal hardware.
- The hard part is engineering: packaging, speed, and knowing when local inference is the right tool.
I want to work on software where on-device inference is a real option, not a demo that only works in a notebook.
03
Education & Skills
B.Sc. Software Engineering
2022 to present
Algorithms and systems in class; local ML and inference experiments on my own time.
- Data structures, OOP, and software design in coursework
- Public GitHub projects: web apps, games, and experiments
- Hands-on with local models, not just API wrappers
Toolkit
- Python Scripts, small ML experiments, tooling around local models.
- TypeScript Full-stack apps (e.g. ttgod-scheduler).
- Ollama / llama.cpp Running and comparing open-weight models locally.
- PyTorch Training loops and evaluation on smaller setups.
- Godot / GDScript Cinderfall: gameplay and architecture.
- Docker Deploying app stacks.
- Git Every project, every iteration.
Currently exploring
- Benchmarking open models with Ollama and llama.cpp
- Quantized weights (GGUF, ONNX) and what you trade for speed
- Small fine-tunes when a general model is close but not quite there
- When local beats cloud on latency and cost, and when it does not
04
Projects
2 repos so far at github.com/SamiulH25, with room for more. Each card is a project I've actually spent time on.