Skip to content
View NikosMav's full-sized avatar

Organizations

@withlope

Block or report NikosMav

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.

Content in all repositories owned by your account will be closed.
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
NikosMav/README.md

Nikos Mavrapidis

Software engineer in Athens, working across backend systems, Linux, automation, and applied AI.

BSc in Informatics and Telecommunications, National and Kapodistrian University of Athens (NKUA), 2017–2022.

Personal site · LinkedIn

Retrieval & recommender work

  • Netflix Catalog Search: BM25, dense, hybrid RRF and cross-encoder reranking over a Netflix title catalog; hybrid + rerank reaches nDCG@10 0.7185 on 28 author-labeled queries.
  • SciFact Hybrid Retrieval: BM25, MiniLM and hybrid RRF on BEIR SciFact test; BM25 nDCG@10 0.676 against the published Anserini BM25 baseline of 0.679.
  • MovieLens Recommender: staged recommender on MovieLens with a bootstrap-CI evaluation harness; item–item kNN reaches NDCG@10 0.120 on ml-1m.

Selected work

  • Lope — Co-founder, part-time alongside SoftCom (Oct 2025 – Sep 2026, paused). AI-native recruitment CRM built by a team of three and run in production with a recruiting-agency design partner: explainable vector search, LinkedIn enrichment, interview intelligence, and an MCP server (1,500+ interviews, ~800 AI-sourced candidates). Case study
  • Camelia Wellness Center — bilingual wellness site: product design, localization, GCP and Cloudflare. Live
  • Worthify — team hackathon for used-car valuation on ~90k listings (R² ≈ 0.93, ~10% median error; Nov 2023 – May 2024). Prototype is no longer live.
  • Database File Structures: heap files, hash files with overflow chains, and a B+ tree in C. Disk blocks, record storage, splits, and scans, with a reproducible build and regression tests.
  • MNIST Vector Search: approximate nearest-neighbour search and clustering in C++, plus autoencoder latent-space search. Includes full-dataset benchmarks.
  • Ubiquitous Computing thesis: BSc thesis (NKUA, published Oct 2023); museum exhibit concept, combining desktop scrollytelling, an optional reading version and browser-based computer vision. PDF

This profile contains personal projects, public case studies, tools, and selected academic work.

Stack

Python · Java / Spring · C/C++ · Robot Framework · Linux · GitLab CI / Jenkins · embeddings / RAG

Pinned Loading

  1. lope-case-study lope-case-study Public

    Case study: Lope, an AI-native recruitment CRM that turns a role brief into a ranked, explainable shortlist (2025-2026)

    HTML

  2. netflix-catalog-search netflix-catalog-search Public

    Offline Netflix catalog search with BM25, dense retrieval, hybrid RRF, cross-encoder reranking, and reproducible relevance evaluation.

    Jupyter Notebook 7 5

  3. scifact-hybrid-retrieval scifact-hybrid-retrieval Public

    SciFact document ranking with BM25, MiniLM, and hybrid RRF, next to the BEIR BM25 anchor. ISOT is the closed-corpus demo, not the evaluation.

    Python 2

  4. camelia-wellness-case-study camelia-wellness-case-study Public

    Product design, bilingual UX and GCP delivery case study for Camelia Wellness Center

  5. worthify-case-study worthify-case-study Public

    AI-driven used-car valuation hackathon case study

  6. maritime-optimization-case-study maritime-optimization-case-study Public

    Portfolio case study for a maritime fuel-mix, FuelEU and EU ETS optimization prototype