Full-Stack & AI Systems Engineer | React 19, TypeScript, Next.js | Python, FastAPI | Microsoft Open Source Contributor
I architect high-performance client-side web applications, standalone developer tools, and hybrid Retrieval-Augmented Generation (RAG) pipelines with deterministic grounding and AST codebase indexing.
- 🏆 Official Microsoft Open Source Contributor — Contributed zero-cloud local hybrid RAG recipes to microsoft/PhiCookBook leveraging
phi-4-miniand SQLite FTS5 (Merged in PR #571). - 📦 GitHub Developer Program Member — Author of Hybrid RAG Issue & PR Assistant, an official marketplace GitHub Action for deterministic lexical and dense semantic repository triage.
- 🤖 AI Intern @ Microsoft (AI Innovators Summer Program) — Engineered local RAG systems using Microsoft Foundry Local SDK, dense vector embeddings, SQLite FTS5 BM25 hybrid retrieval (RRF), and real-time SSE streaming.
- ⚡ Ex-Frontend Intern @ TurkNet — Developed and optimized production user interfaces handling live customer traffic with React, Next.js, TypeScript, and Redux Toolkit.
- 🧩 Systems Architecture & Performance — Specialized in client-side telemetry, Web Workers concurrency, AST lexing, standalone
@vercel/nccpackaging, and on-device SLM inference. - 🏅 Leadership & Structured Problem Solving — Graduate of McKinsey.org Forward Program (MECE framework & strategic communication) & Global Finalist at Aspire Leaders Program (Harvard-founded).
- 🎓 Education — Web Design & Coding at Istanbul University | Computer Programming Graduate from Istanbul Beykent University (3.61/4.00 GPA).
- 📖 Medium Deep-Dive: I Ditched Cloud Vector Databases for SQLite FTS5, and My RAG Pipeline Got 10x Better — Architectural analysis of semantic failure modes in code search, BM25 + Dense embedding trade-offs, and Reciprocal Rank Fusion (
$k=60$ ) mathematics. - 📖 DEV.to Edition: I Ditched Cloud Vector Databases for SQLite FTS5 — Developer-oriented breakdown on eliminating cloud database costs and latency using embedded SQLite FTS5 in RAG pipelines.
- 💬 GitHub Community Discussion #205923 — Technical deep-dive on deterministic BM25 and dense semantic fusion for repository issue and PR triage.
| Domain | Technologies & Tools |
|---|---|
| Frontend & Systems Architecture | React 19, Next.js (App Router), TypeScript (Strict), Three.js (WebGL), Redux Toolkit, Web Workers, Vite, Tailwind CSS, PWA, Vitest |
| Backend & Databases | Python, FastAPI, Node.js (v20/v22/v24), Express, RESTful APIs, SQLite (FTS5 BM25), PostgreSQL, MongoDB |
| AI & Retrieval Systems | Hybrid RAG, Okapi BM25 ( |
| DevOps & Static Analysis | GitHub Actions SDK, AST Static Lexing, Tarjan SCC Graph Algorithms, @vercel/ncc Bundling, Multi-Repo CI/CD Matrix, Docker, Git |
GitHub Marketplace • Node.js 24 • @vercel/ncc • Okapi BM25 • Dense Vectors • RRF Fusion • Phi-4
- Production-Grade GitHub Action: Automates repository triage using a hybrid RAG pipeline fusing Okapi BM25 keyword scoring with dense semantic embeddings via Reciprocal Rank Fusion (
$k=60$ ).- Deterministic Line-Span Grounding: Implements markdown AST-aware chunking to output verified citations (
[file#L<start>-L<end>]), packaged into a single zero-dependency bundle via@vercel/ncc.- Autonomous Sunday Sentinel: Features cloud cron automation (
ecosystem-sentinel.yml) verifying multi-framework compatibility across Python 3.11/3.12 with automated issue triage.- Official Marketplace Listing: Hybrid RAG Issue & PR Assistant
JavaScript (ESM) • Node.js Native • Three.js WebGL • Tarjan SCC • MITRE CWE Sentry • Zero-Dependency • Air-Gapped
- Interactive 3D Architectural Topology: Statically analyzes JavaScript/TypeScript AST dependencies and maps modular graph topology into an interactive 3D spatial environment modeled after the Istanbul Bosphorus.
- Tarjan SCC Cycle Solver & Contract Extractor: Evaluates circular dependency chains in
$O(V+E)$ time with zero external npm dependencies, synthesizing decoupled TypeScript contract interfaces (types/*.contract.ts) and standard Unified Git Diffs.- Security Boundary Sentry & CI Gatekeeper: Audits architectural boundaries against MITRE CWE standards (CWE-668, CWE-200, CWE-798). Operates as a headless CI gatekeeper (
--fail-on-cycle --fail-on-leak) with automated port collision fallback and standalone HTML report generation.- Local-First & Air-Gapped: 100% client-side memory execution with zero cloud telemetry, bundled local Three.js r128, and local WOFF2 variable fonts. Verified by 48 automated native unit and integration tests.
- Live Demo: https://cagrik34.github.io/zenith-istanbul/
React 19 • TypeScript 5.8 • Vite 6 • Web Workers • AST Parser • SQLite FTS5 • Vitest • Zero-Cloud
- Client-Side Isolated Architecture: Developer platform featuring RepoSense (real-time AST codebase topology at 120 FPS), DevForge (JSON-to-TS/Zod, cURL translator, WASM sandbox), and MindVault (sub-5ms SQLite FTS5 BM25 note search).
- Fault-Tolerant Resilience: Production React 19 Error Boundary architecture, sub-vendor Rollup chunking (17.9 kB main gzip), and 100% automated Vitest unit & benchmark coverage.
- Live Demo: https://cagrik34.github.io/zenith-nexus/
Python • React 18 • TypeScript • FastAPI • Foundry Local SDK • phi-4-mini • SQLite FTS5 • Docker
- Offline Local RAG: Executes
phi-4-mini(3.8B) andqwen3-embedding-0.6b(1024-d) locally with zero external API dependency and zero cloud data egress.- Hybrid Search Engine: Fuses dense embeddings with SQLite FTS5 BM25 using Reciprocal Rank Fusion (
$k=60$ ) with an 89.4% composite quality score across RAG evaluation benchmarks.- Repository: microsoft-foundry-local-rag-assistant
React 19 • TypeScript 5.8 • Vite 6 • Web Workers • PWA • Client-Side Memory Architecture
- Client-Side Financial Telemetry: High-performance browser terminal computing quantitative metrics across 1,051 TEFAS mutual funds and market benchmarks natively in memory.
- Concurrent Execution: Dedicated Web Workers running Black-Litterman, Monte Carlo, and HRP risk models with 60 FPS Canvas rendering and automated 4-page A4 PDF reporting.
- Live Demo: https://cagrik34.github.io/zenith-atlas/
Python • TensorFlow/Keras • OpenCV • Tkinter • SQLite
- End-to-end computer vision desktop application utilizing dynamic thresholding to classify 13 distinct emotional intensity levels in real time.
- Modular architecture with live Tkinter GUI telemetry and SQLite session data logging for adaptive UI/UX feedback.
- 🏆 microsoft/PhiCookBook #571 — Merged Official Contributor: Contributed a zero-cloud local hybrid RAG recipe combining SQLite FTS5 BM25 lexical search with
phi-4-minilocal inference via the Microsoft Foundry Local SDK.
- 🌟 Microsoft Türkiye AI Innovators Summer Program (2026) — Selected for hands-on AI architecture & Copilot orchestration.
- 🎓 Aspire Leaders Program (2026) — Global Finalist (Selected among 10,588 finalists from 50,284 worldwide applicants) & Official CEO Recognition.
- 📜 McKinsey.org Forward Program (2026) — 100% Completion (Structured Problem Solving, MECE Framework, Strategic Communication).
- 🌐 Google Developer Groups (GDG) Build With AI Türkiye & HUAWEI Data Science & ML Bootcamp participant.

