Skip to content
View AStroCvijo's full-sized avatar

Block or report AStroCvijo

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
AStroCvijo/README.md

👋 Hey, I'm Jovan Cvijanović

LinkedIn Email CV


🎯 About Me

💼 Applied Scientist at Microsoft
🎓 Software Engineering student at Faculty of Technical Sciences, Novi Sad, Serbia
🏆 1st place at Garaža Frontier Hackathon 2025 · 1st place at National Olympiad in Informatics 2022
🌊 Attended Mediterranean ML Summer School (M2L) 2025


🛠️ Interesting projects

Comparative analysis of monocular depth estimation methods, with robustness evaluation under degraded conditions

Tech: Python, PyTorch, Stable Diffusion, I-JEPA, DepthAnything V2
Highlights: Compared five approaches on NYU Depth V2 (ResNet-50 baseline, frozen SD & I-JEPA features, their fusion, and DepthAnything V2 as SOTA reference), and measured robustness under fog, blur, and low-light.


🔍 RAG Systems — Research to Deployment

From codebase Q&A experiments to a deployed personal RAG assistant

Tech: Python, LangChain, PostgreSQL, pgvector, Ollama, FastAPI
Highlights: CodeRAG — improved Recall@10 from 63.7% → 88.8% via LLM code summaries, enriched metadata, and chunking/embedding experiments. Otto RAG — productionized those learnings into a deployed RAG over Gmail & Google Drive with per-user isolation and OAuth2. Shared agentic pipeline with query rewriting, relevance grading, and hallucination checking.


1st place, Garaža Frontier Hackathon — transform room videos into 3D scenes and swap furniture with AI

Tech: React, FastAPI, Three.js, Depth Anything V3, YOLOv8, Google Gemini
Highlights: Video-to-3D point cloud reconstruction, AI furniture detection with semantic product search, photorealistic replacement generation.


CBOW & Skip-Gram with Negative Sampling from scratch in pure NumPy

Tech: Python, NumPy
Results: 44.2% Google analogy accuracy (CBOW), WordSim-353 ρ = 0.731 (SGNS) on text8. Includes interactive REPL, W&B sweeps, pretrained embeddings, and pytest suite.


Interactive web app for graph visualization and manipulation

Tech: Python, Django, D3.js, JavaScript
Highlights: Plugin architecture for data sources (JSON, XML) and visualizer backends. Multiple views (Main, Tree, Bird/minimap), CLI terminal for graph CRUD.


Full-stack Uber-like application with real-time tracking

Tech: Java, Spring Boot, Angular, PostgreSQL, WebSocket, Leaflet, JWT
Highlights: Live map tracking, WebSocket chat, JWT auth with role-based access, Android companion app, E2E tests with Selenium.


LinkedIn Email

Pinned Loading

  1. cyclops cyclops Public

    Comparative analysis of monocular depth estimation methods (ResNet-50, frozen Stable Diffusion UNet, I-JEPA, SD+I-JEPA fusion, DepthAnything V2) with robustness evaluation under fog, blur, and low-…

    Jupyter Notebook 1

  2. word2vec word2vec Public

    From-scratch NumPy implementation of Word2Vec (CBOW & SGNS) with benchmark evaluation, hyperparameter sweeps, and pretrained embeddings on text8.

    Python

  3. deepgrad/epiplar.io deepgrad/epiplar.io Public

    because rooms are easier to understand in 3d! 👁️

    TypeScript 1

  4. coderag coderag Public

    Retrieval-Augmented Generation (RAG) system over a code repository for a question-answering task

    Python 3

  5. kzi-nastava/mrs-team28-Lavugio kzi-nastava/mrs-team28-Lavugio Public

    Lavugio is a full-stack ride-sharing platform inspired by Uber and Bolt, built as a university project. It features real-time driver tracking, in-app chat, instant and scheduled ride booking, a dri…

    Java 2

  6. power_consumption_prediction power_consumption_prediction Public

    Future power consumption prediction using LSTM, GRU and Transformer models

    Python 13