Project Roulette / Projects

Data & ML and Systems projects

8 of 30 projects. Each has a stack, five features to ship, a difficulty, a time estimate, and a bonus challenge.

Data & ML · Intermediate · 1–2 weeks

Resume ↔ Job Description Matcher

Upload a resume and a job posting. It scores how well they match, highlights missing skills, and suggests stronger bullet points.

  • Python
  • sentence-transformers
  • spaCy
  • FastAPI
  • Streamlit

Features

  1. PDF parsing and section extraction (experience, skills, education)
  2. Skill extraction against a curated skills taxonomy
  3. Semantic similarity scoring with a readable explanation
  4. Gap report listing missing and weak skills
  5. A small labeled test set that measures accuracy

Bonus: Benchmark three embedding models on your test set and publish the results table in the README.

Data & ML · Intermediate · 1–2 weeks

Webcam Hand-Gesture Controller

Recognizes hand gestures from a webcam in real time and uses them to control slides, media playback, or smart lights.

  • Python
  • MediaPipe
  • PyTorch or scikit-learn
  • OpenCV

Features

  1. Hand landmark extraction from live video
  2. Data-collection tool for recording custom gestures
  3. Trained classifier with a confusion matrix and accuracy report
  4. Smoothing and debounce so actions don't misfire
  5. Config file that maps gestures to actions

Bonus: Export the model to ONNX and run it fully in the browser with a live demo page.

Data & ML · Advanced · 2–3 weeks

Predictive Maintenance for Jet Engines

Predicts the remaining useful life of jet engines from sensor time series using NASA's C-MAPSS dataset, served behind an API with a monitoring dashboard.

  • Python
  • pandas
  • LightGBM or PyTorch
  • MLflow
  • FastAPI
  • Docker

Features

  1. Exploratory analysis and time-series feature engineering
  2. A baseline model and an LSTM, compared with proper time-based validation
  3. Experiment tracking in MLflow
  4. Model served through FastAPI inside Docker
  5. Dashboard showing predictions per engine

Bonus: Data-drift detection that flags when incoming sensor data stops looking like the training data.

Data & ML · Intermediate · 2 weeks

Codebase Q&A Assistant

Index any Git repository and ask questions like "where is authentication handled?" Answers cite the exact files and lines they came from.

  • Python
  • tree-sitter
  • sentence-transformers
  • pgvector or Qdrant
  • FastAPI
  • An LLM API

Features

  1. Code-aware chunking by function and class using tree-sitter
  2. Hybrid keyword and vector search
  3. Answers with file and line citations
  4. Incremental re-indexing when new commits land
  5. An evaluation set of 30+ questions that measures retrieval accuracy

Bonus: A VS Code extension so you can ask questions without leaving the editor.

Systems · Advanced · 2–3 weeks

LSM-Tree Key-Value Store

A persistent key-value database built from scratch with a write-ahead log, memtable, SSTables, and compaction, like a tiny RocksDB.

  • Rust or C++
  • TCP
  • criterion / Google Benchmark

Features

  1. Write-ahead log with crash recovery
  2. Memtable flushed to sorted SSTable files
  3. Bloom filters to skip unnecessary disk reads
  4. Background compaction
  5. Simple TCP protocol with a CLI client

Bonus: Benchmarks against SQLite with throughput and latency graphs in the README.

Systems · Advanced · 2–3 weeks

RISC-V Emulator with Debugger

An RV32I CPU emulator that runs real programs compiled with GCC, plus a step debugger for registers and memory.

  • C++ or Rust
  • RISC-V GCC toolchain
  • SDL2 (optional)

Features

  1. Fetch, decode, and execute for the full RV32I instruction set
  2. ELF loader for compiled binaries
  3. Memory-mapped UART so programs can print
  4. Step debugger with register and memory views
  5. Passes the official riscv-tests suite

Bonus: A 5-stage pipeline view that visualizes hazards, stalls, and forwarding cycle by cycle.

Systems · Advanced · 2–3 weeks

Mini Container Runtime

Your own version of docker run, built from scratch: isolated processes with their own filesystem, process tree, network, and resource limits.

  • Go or C
  • Linux namespaces
  • cgroups v2
  • OverlayFS

Features

  1. PID, mount, UTS, and network namespaces
  2. Root filesystem set up with pivot_root from an image tarball
  3. CPU and memory limits using cgroups v2
  4. Layered filesystems with OverlayFS
  5. veth networking so containers can reach the internet

Bonus: Pull and run real images from Docker Hub by implementing the registry API.

Systems · Advanced · 2–3 weeks

Compiler for a Small Language

A compiler for your own small language, with variables, functions, loops, and structs, that outputs WebAssembly or x86-64 assembly and runs real programs.

  • Rust, C++, or OCaml
  • WebAssembly (WAT) or x86-64
  • A test harness

Features

  1. Hand-written lexer and recursive-descent parser
  2. Type checker with clear error messages that point at the source
  3. An intermediate representation with constant folding and dead-code elimination
  4. Code generation to WebAssembly or assembly
  5. Test suite of example programs with expected output

Bonus: A web playground that compiles and runs programs right in the browser.