Skip to content

300 Days Of Code - Year 2 (2025)

Learning Journey

Welcome back to my coding journey! This repository marks the second year of my 300 Days of Code challenge.

Last year focused heavily on daily logs and updates within the realm of vision technologies. This year, the focus shifts towards deeper dives into key bootcamps and substantial projects, moving away from daily standup-style logging.

The goal remains consistent: dedicated coding effort. However, the definition of a "minimum viable day" is now centered on completing the LeetCode daily challenge. This ensures consistent problem-solving practice while allowing for more flexibility to concentrate on larger learning modules and projects.

Challenge Structure

  • LeetCode: Daily challenges
  • Bootcamps: Each bootcamp tracked as a submodule
  • Projects: Larger projects get their own folders
  • Milestones: Progress logged via sprint highlights & project completions, not daily standups

Strategic Learning Phases

Instead of tracking 365 individual days, I'm organizing the year into strategic learning phases:

Phase Primary Focus Timeline Key Learning & Projects
1 Generative AI Jan–Apr Completed GenAI Bootcamp · Built multi-agent chatbots, RAG demos, ASR/TTS experiments
2 MCP & Multi-Agent Systems May–Jul AWS AI Practitioner cert · Networking Bootcamp (multi-VPC design) · Agentic Comic Generator (Hugging Face Hackathon) · LLM Zoomcamp · 100DaysOfCloud Discord Bot · Discord MCP Server · Anthropic MCP
3 MCP Tooling & Cloud Aug–Oct Albumentations MCP Server (Kiro Hackathon) · Discord MCP Server · Albumentations MCP VLM feature (Nano Banana Hackathon) · AWS SAA (BeSA) prep · Korean B2 (KSI) · MLOps Zoomcamp · LLM Zoomcamp
4 Cloud Practice & Consolidation Nov–Dec 100 Days of Cloud (AWS SkillBuilder + projects) · Kiroween Hackathon (skeleton structure track) · EdgeMind paper finalization · Backlog cleanup (closing open repos)
Note

Some tracks (e.g. certifications, Zoomcamps, MCP tools) span multiple phases. Phases are for showcasing emphasis.

Projects in Progress

100 Days of Cloud (Nov–Dec 2025)

Daily AWS practice log exploring networking, IAM, containers, and GenAI on the cloud. Goal: strengthen AWS fundamentals while applying AI services in real workflows.

Project Repo →

LangGraph Mini Project (Planned)

Short RAG-based build to showcase agent orchestration with LangChain + LangGraph. Goal: add one polished, industry-relevant LLM project for job readiness.

Completed Projects & Outcomes

Major Achievements

Yoga Assistant Knowledge RAG (Nov 2025)

End-to-end retrieval-augmented generation system for yoga knowledge with rigorous evaluation and production deployment.

  • Achieved 23% MRR improvement through hybrid retrieval (BM25 + vector search)
  • Benchmarked 5 LLMs with 3 prompt templates (15 combinations) on ground truth questions
  • Selected DeepSeek-V3 for 90% relevant answer rate
  • Demonstrated that query rewriting and re-ranking can degrade performance
  • Production deployment with PostgreSQL logging and Grafana monitoring (7 visualization panels)

Tech stack: Python, BM25, sentence-transformers, DeepSeek-V3, PostgreSQL, Grafana, Docker

→ Project Repo

Dynamic Knowledge Graph Agents (Nov 2025)

Multi-agent system that transforms unstructured text into structured, visual knowledge graphs using Google ADK and code execution.

  • Sequential agent pipeline: ResearchAgent → OntologyAgent → VisualizationAgent
  • Extracts structured triplets (Subject → Predicate → Object) from natural language
  • Generates interactive visualizations with multiple layout algorithms (Barnes Hut, Force Atlas 2, Hierarchical)
  • Custom observability plugin for metrics tracking
  • Streamlit frontend with FastAPI backend

Tech stack: Google ADK, Gemini 2.5 Flash, NetworkX, Matplotlib, FastAPI, Streamlit

→ Project Repo

EdgeMind — AWS Agents Hackathon (Oct 2025)

5G–6G edge orchestration framework integrating multi-agent swarms with Strands Agents and MEC for sub-100 ms decisioning.

  • Built during the AWS Agents Hackathon; now evolving into a research-backed framework for agentic orchestration at the network edge
  • Implements structured agent memory, ReAct-style critique loops, and JSON-based orchestration
  • Serves as foundation for an ongoing paper on 6G Edge Intelligence

Project Repo → | Live Demo →

BeSA Bootcamp — Migration & Modernization on AWS (Sep–Oct 2025)

Focused on large-scale migration patterns and modernization strategies across hybrid environments. Completed full curriculum including migration waves, VMware Cloud on AWS, and modernization architectures.

Project Repo →

Discord MCP (Proof of Concept)

A modular, MCP-compliant Discord automation server that bridges AI assistants (Claude, etc.) with Discord for agent-tool orchestration, role-based automation, and workflow management.

Currently in beta testing, consolidating from 26 tools to 9 production-ready modules.

  • Implements atomic + workflow-based tools (discord_send, discord_discovery, discord_campaign, etc.)
  • Designed for AI assistants to manage communities, run campaigns, and automate Discord workflows
  • Features structured SQLite memory, natural-language intent parsing, and Jinja2 templated messaging
  • Next phase: optimizing API exposure and formalizing pipeline tests for production readiness

Project Repo →

Albumentations MCP (Kiro Hackathon, Jul - Sept 2025, Nano Banana Hackathon Sept 2025)

MCP-compliant image augmentation tool using Albumentations. Built to enable safe, auditable augmentation pipelines driven by natural language prompts. Integrated with Nano Banana (Gemini 2.5 VLM) for hybrid image-to-image workflows.

  • Natural language → structured transforms ("add blur and rotate 15 degrees" → GaussianBlur + Rotate)
  • Full MCP protocol support with tools (augment_image, validate_prompt, list_available_transforms, etc.)
  • Preset pipelines for segmentation, portrait, and lowlight augmentation
  • Deterministic seeding and structured session logging for reproducibility
  • VLM integration: Nano Banana/Gemini for vision-language guided edits

Tech stack: Python, Albumentations, MCP, Gemini/Nano Banana, uvx, Kiro IDE

Project Repo →

Hugging Face Hackathon (Jun 2025)

Agentic Comic Generator - Hugging Face / LlamaIndex Hackathon submission. A multi-agent comic generation system powered by GPT-4o, SDXL, and LlamaIndex.

  • Dual-agent setup: Agent Brown (reasoning + critique) and Agent Bayko (image generation via Modal)
  • Custom LlamaIndex event system with persistent memory and structured state logging
  • Modal serverless endpoints for SDXL generation
  • Gradio UI with real-time agent feedback

Tech stack: GPT-4o, Mistral Codestral, Hugging Face Diffusers, Modal, Gradio

Project Repo →

GenAI Bootcamp (Feb–Apr 2025)

Achievement
  • Completed 100% of all project submissions
  • Awarded RED SQUAD recognition - highest grade awarded
  • Focus Areas: TOGAF, ASR, TTS, AI assisted Language Learning, Agents, RAG, MUD games, handwriting recognition, multi-agent chatbots
Challenges Faced
  • Struggled with frontend/backend integration
  • Major friction with deployment (Docker, Spaces, Firewalls)
  • Learned the importance of understanding cloud tools earlier in project cycles

LLM Zoomcamp (DataTalksClub) - Partial (Jul–Oct 2025)

Completed core modules on embeddings, vector databases, and evaluation. Capstone paused due to concurrent hackathons — planned revisit for early 2026.

Course Link →

AWS Learning Highlights

LeetCode Monthly Challenge Badges

Jan Feb Mar Apr May Jun
✓ ✓ ✓ ✓ ✓ ✓
Jul Aug Sep Oct Nov Dec
✓ ✓ - ✓ - -

Other Certifications

  • Introduction to Model Context Protocol
  • Model Context Protocol: Advanced Topics
  • MCP: Build Rich-Context AI Apps with Anthropic
  • Practical Multi AI Agents and Advanced Use Cases with crewAI
  • Korean Introductory Course KSI
  • MLOps Zoomcamp

Research Papers

Publications & Work in Progress
  • "EdgeMind: Multi-Agent Orchestration in the Transition from 5G to 6G" → in progress
  • "Turn Detection in AI Language Learning" → Dissertation Link
  • "Beyond Human Mimicry: Rethinking AI Identity" → in progress

Exploratory Topics

Research Directions

CFD Memory - Investigating fluid-inspired memory systems where contextual relevance emerges from dynamic flow patterns rather than static vector weights. Focus: memory decay, attention turbulence, and mesh-based semantic diffusion.

Korea as an AI Testbed - Exploring South Korea's unique role as a contained ATS-friendly deployment zone for multilingual AI tooling. Emphasis: linguistic modeling, policy alignment, and interface testing across hyper-digital infrastructure.

Repository Structure

  • leetcode/ → LeetCode Daily Solutions
  • certs/ → Certification images and PDFs
  • bootcamps/ → GenAI, Networking, and other bootcamp repos
  • courses/ → Self-paced course materials

Connect & Collaborate

I believe in learning in public and welcome collaboration. If you're interested in any of the projects or want to pair program on anything, you can reach out to me via LinkedIn.

← Back to Blog