GEORGE CHKHAIDZE

Competitive Programming โ€ข AI Research โ€ข Full-Stack Development

Carnegie Mellon University Qatar CS Junior | ICPC World Finals 2025 Finalist | Full-Stack

Contact Codeforces GitHub
๐Ÿ† ICPC World Finals 2025
Carnegie Mellon Qatar

ACADEMIC & PROFESSIONAL JOURNEY

Experiences that shaped my engineering skills.

2023-2027
Carnegie Mellon University Qatar

Bachelor of Science in Computer Science, graduating 2027. Selected courses: Applied Software Engineering, Parallel & Sequential Data Structures, Great Theoretical Ideas in CS, Browser Security.

2025
ICPC World Finals - Azerbaijan

Represented Carnegie Mellon University Qatar at the International Collegiate Programming Contest World Finals, competing among the world's elite programming teams.

2024
QCPC Champions & ACPC Top 10

Won 1st place at Qatar Collegiate Programming Contest and ranked 10th out of 112 teams at Arab & Africa Collegiate Programming Championship in Egypt.

2023
Scientific Committee & Problem Setter

Served as Scientific Committee Member and Problem Setter for European Junior Olympiad in Informatics.

2022
IOI 2022 - Indonesia

Represented Georgia at the 34th International Olympiad in Informatics in Indonesia, competing among the world's top young programmers.

2021
APIO Bronze Medal

Earned Bronze medal at Asia-Pacific Informatics Olympiad, demonstrating exceptional algorithmic thinking and problem-solving skills.

TECHNICAL EXPERTISE

Tools & Skills that I learned when developing useful software.

โšก

C++

STL & algorithm optimization. 2200+ problems solved on Codeforces.

๐Ÿ

Python

Reflex framework, Django, Flask, FastAPI, NumPy.

โ˜•

Java & Scala

Functional and object-oriented design for concurrent backend systems.

๐Ÿš€

JavaScript

React with Vite, custom hooks, TailwindCSS, production deployment.

๐Ÿงช

Testing & DevOps

Vitest, comprehensive test coverage, CI/CD pipelines, code obfuscation.

๐Ÿค–

AI & Machine Learning

LLaMA 3 fine-tuning, BERT models, Ollama, similarity matching algorithms.

๐Ÿ—„๏ธ

Database & Backend

PostgreSQL, Supabase, MongoDB, Express.js, Node.js, authentication systems.

PERSONAL PROJECTS

Some projects I worked on recently.

ChessMaxx

Platform delivering personalized chess opening training exercises.

Live Full-stack build Check it out
Professional chess opening training platform with interactive board play, 3000+ opening variations and personalized statistics.

Architecture & Engineering

  • Clean layering: Domain/Application/Infrastructure separation with explicit contracts
  • Deployment pipeline: Automated obfuscation, minification, and cache-busting releases
  • Performance focus: Lazy loading, memoized selectors, and static asset optimization

Interactive Features & User Experience

  • Engine: Chess.js validates moves, generates lines, and scores positions
  • UI: Drag-and-drop interactions with motion feedback and progress cues
  • Insight dashboards: Real-time stats, analytics, and skill trend visualizations
  • Authentication: Google OAuth with session and role-aware access
React & TailwindCSS Chess.js Supabase Google OAuth

Reconnect

Community-Driven Lost & Found Platform.

Full-stack build Hackathon finalist
Social platform for lost-and-found coordination and collaborative missions to keep volunteers aligned until every case is closed.

Innovation & Community Impact

  • Recognized build: Finalist among 50+ teams at CMUQ Lifelines Hackathon for social resilience
  • Threaded collaboration: Every report creates a dedicated channel with tagging, roles, and upvotes
  • Geolocation: Map pins reveal activity clusters
  • Knowledge archive: Playbooks and success stories for future recovery missions

Technical Architecture & Collaboration Tools

  • Structured data core: PostgreSQL schema for users, media assets, case updates, and communities
  • Real-time comms: Live chat plus AI assistant accelerate coordination and triage
  • Alerting fabric: Event-driven emails and notifications for matches and verified sightings
  • Responsive delivery: Mobile UX ensures field teams stay productive anywhere
Reflex PostgreSQL Geolocation API Mobile Responsive

ACADEMIC PROJECTS

Projects I did under university courses.

FileStack - Distributed File System

Custom RMI stack built from the ground up for CMU 15-440.

From-scratch RMI Multi-threaded core
Implemented a distributed file system that speaks our own Remote Method Invocation protocol. Dynamic proxies, serialization, and replication guarantees deliver file access across networked storage nodes.

Core Distributed Systems Capabilities

  • Complete operations: CRUD support with duplicate detection and automatic directory management
  • Self-registering nodes: Storage servers discover and sync with the naming service on boot
  • Global namespace: Bidirectional path mapping keeps remote storage transparent to clients
  • Reliability: Thread-safe logic and graceful failure handling across the cluster

Custom RMI Architecture & Implementation

  • Protocol authoring: Java dynamic proxies + TCP sockets + bespoke serialization pipeline
  • Concurrency: Worker threads prevent blocking and preserve throughput
  • Three-tier topology: Clients, naming server, and storage nodes coordinate via RPC
  • Intelligent dispatch: Transparent routing between local and remote method calls
Java RMI Distributed Systems Multi-threading TCP Sockets Dynamic Proxies Concurrency File Systems

BERT-based XSS Detection System

Browser security research project for CMU 17-403.

IEEE-format study Real-time scanning
Created a defensive ML system that flags cross-site scripting payloads before they land in production. Dataset balancing, augmentation, and latency tuning make it practical for real-world web traffic.

Research & Results

  • Accuracy: 94.74% F1 / 93% zero-shot on community payloads
  • Bias mitigation:Custom augmentation removes over-reliance
  • Corpus: 20,070 labeled samples spanning comments, form inputs, and adversarial payloads
  • Production viability: <100ms inference latency keeps the browser experience snappy
  • Publication-ready: Full IEEE paper with methodology, and literature synthesis

Technical Architecture

  • Model design: Fine-tuned BERT-base-uncased with task-specific classification head
  • Payload diversification: Generated function call variants to stress-test the classifier
  • Hardening tests: Validated against public exploit repos to confirm generalization
  • Deployment path: GPU-accelerated inference service with CPU fallbacks and batching
BERT Fine-tuning PyTorch Data Augmentation Web Security XSS Detection Browser Security

ENTERPRISE PROJECTS

Some projects I did with/for enterprises.

Eclipse OpenJ9 AI Integration

Enterprise AI triager shipped with Microsoft & IBM mentors; merged into the JVM's production workflow.

Merged contribution Issue triaging AI
Led an AI issue triager for Eclipse OpenJ9 issues, pairing LLaMA 3 fine-tuning with automation pipelines. The system reduces manual triage for the foundation.

Enterprise Impact & Mentorship

  • Executive backing: Mentored by Microsoft Azure CTO Office, Red Hat leadership, and IBM engineers
  • Production launch: AI workflow merged and supporting the Eclipse developer community
  • Operational architecture: MLflow experiments, Azure VM deployments, and containerized pipelines

Technical Leadership & Contributions

  • Data validation: Logistic regression to ensure balanced training signals for Triager
  • AI infrastructure: Designed ChatGPT-Gemini server stack with observability hooks
  • Workflow automation: Used GitHub Actions for intelligent summaries and communication gating
  • Conversational tooling: Built Slack co-pilot delivering real-time issue insights to maintainers
LLaMA 3 Fine-tuning MLflow Azure Deployment Docker Slack API GitHub Actions Data Balancing

Ooredoo B2B Assistant - Agentic RAG

Bilingual enterprise assistant orchestrating six parallel tools to resolve complex customer operations.

Agentic RAG Hackathon Finalist
Designed a concurrent FastAPI server where AI reasoning agent FAQ retrieval, CRM lookups, maintenance, and more. Responds for Ooredoo's B2B customers in English & Arabic.

Production-Ready Features

  • LLM flexibility: Supports OpenAI GPT-4o-mini or self-hosted Ollama models with Redis caching
  • Smart routing: Auto-classifies intents, escalates by SLA, and tracks service requests
  • Security: API keys, PII sanitization, retry logic, rate limiting, and Prometheus metrics

Agentic Synthesis & Advanced Architecture

  • Parallel ReAct: Six specialized tools run concurrently via asyncio.gather
  • Reasoned synthesis: LLM synthesizes tool outputs, resolves conflicts and responds to user
  • Bilingual NLP: Intent detection, identity extraction, and generation in English & Arabic
  • Multi-source retrieval: BM25 FAQs, hybrid product search, SQL CRM, SQLite FTS5 tickets, maintenance APIs
FastAPI LLM (OpenAI/Ollama) RAG Architecture Agentic AI BM25 Retrieval SQLite FTS5 Redis Cache Bilingual NLP

INTERNATIONAL ACHIEVEMENTS

I like winning.

๐Ÿ†

ICPC WF 2025

Finalist - Azerbaijan
Carnegie Mellon University Qatar

๐Ÿฅ‡

QCPC 2024 Winner

1st Place
Qatar Collegiate Programming Contest

๐ŸŒ

ACPC 2024 Top 10

10th out of 112 teams
Arab & Africa Championship - Egypt

๐ŸŒ

IOI 2022

Georgian Representative
34th International Olympiad - Indonesia

๐Ÿฅ‰

APIO 2021 Bronze

Bronze Medal
Asia-Pacific Informatics Olympiad

๐Ÿฅˆ

MIA 2019

2nd Place
Millennium Innovation Award STEM

๐Ÿ’ป

Codeforces

2200+ Problems Solved
190+ Contest Participations

๐Ÿ‘จโ€๐Ÿซ

Olympiad Organizer

Scientific Committee 2023
European Junior OI & Georgian National OI