Case studies, systems notes, and role fit organized as one editorial surface.
Public repo
Neural Network From Scratch
An educational neural network implementation with deterministic evaluation, logs, tests, and a reproducible pipeline.
Why this case matters
This stays public because it shows engineering discipline inside a study project: reproducible evaluation, smoke checks, and structured docs.
Public repo, educational but well-scoped
Inspect first
Start with the repository surface: README, setup path, CI, docs, and the evidence blocks that show the system survives outside the happy path.
Open repositoryContext
Study repositories only help a portfolio when they show craft, not just curiosity. I kept this one because it moved beyond a notebook exercise into a small, testable system with deterministic evaluation and report artifacts.
Problem
I wanted to understand the mechanics of a neural network without hiding behind framework abstractions. At the same time, I wanted the repo to prove that learning projects can still honor engineering standards.
Constraints
- No high-level ML framework.
- Reproducible results for CI and portfolio demos.
- Simple artifacts that can be inspected later.
Core proofDeterministic evaluation script
Repository signalLogs and smoke checks in CI
Operational concernPortable JSON and JSONL artifacts
Architecture
dataset utils
-> normalization and split
-> model
-> backprop training
-> evaluation metrics
-> JSON / JSONL logs
Reproducibility routeData prep, training, deterministic evaluation, and plain-file artifacts
01Normalize a tiny dataset
The data utilities keep preprocessing readable and deterministic so the repo can be rerun the same way in CI and local demos.
02Train without framework shortcuts
Forward pass and backprop stay visible in code, which is the educational point of the repository.
03Evaluate with a fixed seed
The evaluation script locks parameters and seed so portfolio screenshots and CI outputs are reproducible.
04Store artifacts in plain files
JSON and JSONL logs make the result surface portable, inspectable, and easy to discuss without heavier tracking infrastructure.
Decisions and trade-offs
- I used framework-free implementation because the learning value was the point.
- I added a deterministic evaluation script because reproducibility is what separates a study repo from a throwaway experiment.
- I kept logs in JSON and JSONL rather than adding a heavier experiment tracker too early.
What worked
- The repo now communicates rigor instead of just enthusiasm.
- The evaluation script provides an easy conversation path in interviews.
- The CI-oriented smoke evaluation is stronger than a basic
pytest -qalone.
What is still incomplete
- A small benchmark report or confusion matrix export would strengthen the artifact story further.
- There is room for richer experiment configuration without losing the repo's educational focus.
Evidence
Outputs:
- logs/eval-summary.json
- logs/eval-history.jsonl