• Alicedb is incremental in process database library for C++, that works by executing transformations on changes in streaming model.
  • Supports many core operations: projection, filter, union, except, intersect, join, product, distinct, aggregation.
  • Persistent storage and bufferpool with io_uring.
  • Internaly consiset via timestamps.
  • Operating system primitives for the Raspberry Pi 3b+, targeting the ARMv8-A (AArch64) architecture.
  • Virtual memory.
  • Interrupts handling.
  • Process scheduling.
  • Read only file system.
  • Still lacks many importannt stuff like more syscall, signal handling and multicore support.
  • LLMs serving engine.
  • Supported model: qwen2_5_0_5b_instruct.
  • Supported backends: cpu, cuda.
  • Radix kv cache.
  • Chunked prefill, continous batching.
  • Multi process with communication via zmq.
  • Fast api server with GET /v1/models and POST /v1/chat/completions.
  • Distributed consensus algorithm written in GO
  • Replicated state machine.
  • Leader election.
  • Log replication.
  • Cpu only deep learning framework using numpy as backend.
  • Autograd on N-dimensional Matrices.
  • SGD with momentum and Adam optimizers.
  • Dataloader
  • Custom weight initialization.
  • Writen as 3 person, team project for deep learnign course.
  • I was responsible for implementing sampling algorithm based on sciencific papers.
  • Trained model on google cloud using TPU’s

Xi compiler

  • Lexer and parser in ocamllex and menhir.
  • Bidirectional typechecker.
  • Codegenerator translating AST into intermediate low level language.
  • Register allocation algorithm that allow languange to be executed on real architecture with finite amount of registers.
  • Live variable analysis optimalization.

since this was done as university project its not available publicly

Other smaller projects:

  • HTTP Webserver serving static files in pure c
  • Simple unix shell
  • Linked-list based Malloc
  • Traceroute
  • Few linux device drivers
  • Map-reduce algorithm implemenation in go