Full-Stack Engineer & Team Lead

Gözde Şimşek
Full-Stack Engineer & Team Lead · İzmir

I ship agent runtimes

I lead full-stack GenAI delivery — spec before code, review before merge, durable agent loops, MCP integrations, and production RAG.

Seeking Senior/Staff AI Engineer roles — remote or relocation · based in İzmir, Turkey

  • 25 apps in the MCP catalog

    Microsoft 365, Google, Slack, Salesforce, SAP and more — shipped at Skymod

  • 4 model providers

    OpenAI, Anthropic, Gemini and Skymod's in-house LLM behind one gateway

  • 4 talks & posters

    including SKY AI Summit 2026, Istanbul

  • 4 years in production

    React, Node and Python across three companies

Experience 01

Built in production, led in practice.

From frontend dashboards to enterprise GenAI platforms — hands-on delivery with team leadership at Skymod.

Nov 2024 — Present

Skymod

Full-Stack Engineer & Team Lead

İzmir, Turkey

Skymod builds SkyStudio, an enterprise AI platform for assistants, workflows, and business-system integrations.

  • Lead the team building SkyStudio: React/Vite redesign, an Express backend-for-frontend (BFF) in front of the platform's services, multi-tenant workspaces, and per-plan feature flags.
  • Shipped the production agent runtime — durable runs that survive reconnects, a planner that builds tool-call graphs, human-in-the-loop (HITL) approvals, evaluation and tracing, tenant-isolated storage.
  • Built the MCP integration layer — a 25-app connector catalog incl. Microsoft 365, Google, Atlassian, Slack, Salesforce, SAP, HubSpot, GitHub, and Turkish ERP/accounting systems (Canias, Paraşüt) — with an OAuth token vault, workflow nodes, and agent tool calls.
  • Built SkyBuilder, the platform's in-product assistant: it routes each request to the right editor and generates complete workflow graphs from plain-language prompts.
  • Own the workflow and retrieval layer: React Flow editor, event triggers (webhooks, email), Milvus-based RAG (chunking, embedding, OCR, reranking), LangChain orchestration across providers.
  • Delivered a data-analysis agent (FastAPI, sandboxed code execution, streamed outputs) and a model gateway across OpenAI, Anthropic, Gemini, and GOAT — Skymod's in-house LLM — with automatic failover.
Dec 2023 — Aug 2024

Genfoquest Analytica

Part-time Full-Stack Developer

İzmir, Turkey
  • Built scalable React UIs with Hooks and Context API.
  • Developed high-performance FastAPI services with async request handling.
  • Optimized PostgreSQL workflows with indexing and query tuning.
  • Containerized applications with Docker for reliable deployments.
Sept 2022 — July 2024

Uniqgene

Frontend Developer

İzmir, Turkey
  • Designed responsive web and mobile interfaces with React, React Native, Next.js, and TypeScript.
  • Managed storage and databases with Amazon S3, Amazon RDS, MongoDB, and MySQL.
  • Built backend services with Node.js and Express, integrating RESTful APIs.
  • Improved data-fetching with Axios and maintainable state via Context API.

Selected work 02

Case studies, not just screenshots.

Three views of SkyStudio, the enterprise AI platform I lead at Skymod — the platform, workflows & SkyBuilder, and the model/RAG stack.

Selected work 01

SkyStudio

Enterprise AI platform · Skymod

Skymod's enterprise AI platform — assistants, chat, workflows, RAG, and MCP connectors. React/Vite front end; an Express BFF signs every request before it reaches the backend services.

Agent runtime
Durable agent runs with tenant isolation (Postgres row-level security): planner + tool graph, per-step human approval, retrieval and memory in context, evaluation, artifacts on S3, streams that resume after a reconnect.
Platform
Embedded UI pinned to exact build versions, signed request context, per-plan feature gating, Azure AD / Google sign-in, OpenTelemetry tracing across services.
  • Agent runtime
  • HITL
  • MCP
  • Express BFF
  • React
  • Supabase
SkyStudio

Selected work 02

Workflow & SkyBuilder

React Flow · workflow engine

A visual workflow engine plus SkyBuilder — the platform's in-product assistant that routes requests and generates full workflow graphs from natural language.

SkyBuilder
Knows which editor you're in (chat, assistant, workflow) and generates graphs that match the real node catalog, so the result runs as-is.
Execution
Runs LLM, knowledge, logic, and MCP connector nodes — Outlook, Slack, Jira, Confluence, Google Workspace, SharePoint, Salesforce, SAP, HubSpot, GitHub, generic REST, and more. Published flows start from webhooks or incoming email.
  • SkyBuilder
  • React Flow
  • MCP
  • Webhooks
Workflow & SkyBuilder

Selected work 03

Multi-LLM & retrieval

Model gateway · Milvus RAG

One provider catalog (OpenAI, Anthropic, Gemini, and GOAT — Skymod's in-house LLM) with LangChain orchestration, a failover policy, and a Milvus-backed retrieval pipeline.

Models
A central gateway — one integration surface for chat, the agent runtime, and workflow LLM nodes.
RAG
Chunking, embedding, OCR, and reranking services feeding Milvus search for knowledge bases and chat.
  • LangChain
  • Milvus
  • Redis
  • S3
  • Failover
Multi-LLM & retrieval

How I ship 03

See it, ship it, keep it.

My default mental model for agentic systems — plan, tools, trust, trace, and evaluation — whether the surface is a BFF-backed platform, an internal API, or a demo like this.

Agentic loop · live

plantoolsHITLanswer

Planner builds a tool DAG (bounded steps, serial or parallel).

step 1 of 4

Express same-origin BFF · agent proxy

const ctx = signContext(req.session);
 assertEntitlements(ctx, "agentRun");
 proxySSE("/api/agent/stream", ctx);
// never trust tenant IDs from the browser alone
  • Signed session context
  • Server-side entitlement check
  • SSE stream to agent runtime

Trace line 1

chatrequeststream
run = agent.start(turn)
await executeTool(step)
await hitl.approve(step)
return synthesize(run)
run hr_8f2a
step planner
tool —

Name tag

Agent runtime

Planner, tool DAG, retrieval, memory — one durable loop in production.

Evaluation · artifacts

Did the run pass policy + judge?

Trace inspect + S3 artifacts stay workspace-scoped under RLS.

How I work

I ship with GitHub Spec Kit

In Cursor or Claude. I review each artifact before the next step.

  1. /speckit.​specifyspec.md
  2. /speckit.​planplan.md
  3. /speckit.​taskstasks.md
  4. /speckit.​implementPR · CI

Talks 04

On stage & in the room.

Summits, symposium posters, thesis defense, and international COST workshops — mostly GBM imaging and deep learning on histopathology.

Presenting at SKY AI SUMMIT 2026 in Istanbul
2026Speaker

Enterprise AI agents on company data — SkyStudio & Skymod

SKY AI SUMMIT

Bulutistan & SabancıDx Hangar · Istanbul

Event page
HIBIT 2025 poster presentation on GBM morphometric analysis
2025Poster

Predicting Glioblastoma Molecular Subtypes and Clinical Outcomes Through AI-Driven Morphometric Analysis: A Grid-Based Histopathological Approach

HIBIT 2025 — 18th International Symposium on Health Informatics and Bioinformatics

Istanbul Medipol University

Master's thesis defense on GBM MR histopathology prediction
2025Master's thesis

Development of an Artificial Intelligence Model for Predicting Multiple Histopathological Features from MR Images of Glioblastoma (GBM) Patients

Ege University · Bioinformatics (AI)

İzmir, Turkey

Net4Brain COST meeting presentation in Prague
2024Presentation

Deep learning approaches to brain tumor histopathology and gene expression levels

Net4Brain COST Action

Czech Academy of Sciences · Prague, Czech Republic

Net4Brain

Notes 05

How I think about shipping AI.

Principles I apply across GenAI platforms, health-tech products, and full-stack delivery — not tied to one codebase.

Spec

Write it down before the agent codes

I start from specs and plans — a full spec on agent-heavy work, lighter briefs elsewhere — so scope survives handoffs, reviews, and context limits.

Trust

Server-side truth, not UI flags

The browser is a view. I verify auth, tenancy, and entitlements on the server or BFF before side effects — same rule from GenAI platforms to clinical dashboards.

HITL + eval

Approve the step, then judge the outcome

High-impact steps — external APIs, MCP tools, data writes — get a human checkpoint when needed. I evaluate results (rules or review) before they become canonical.

Fit

Not every problem is an agent loop

Sometimes a form, a workflow graph, or plain REST is enough. I pick the smallest durable pattern and keep boundaries explicit instead of forcing one runtime everywhere.

Projects 06

More work you can browse.

Product UIs, clinical tools, lab interfaces, and GenAI surfaces.

Skymod SkyStudio
React · Vite · Express BFF · MCP

Skymod SkyStudio

Skymod's enterprise AI platform — workflows, assistants, MCP connectors, marketplace.

Skymod Workflow
SkyBuilder · React Flow · MCP

Skymod Workflow

SkyBuilder assistant + React Flow canvas — MCP nodes, webhook and email triggers.

Clinical Data Management
React · Node.js

Clinical Data Management

Structured clinical workflows and data capture for research teams.

BMGLab Website
React

BMGLab Website

Lab presence site with a clean scientific product voice.

Uniqgene Website
React

Uniqgene Website

Public product site for genetics-driven health experiences.

Uniqgene Report Dashboard
React · Node

Uniqgene Report Dashboard

Report surfaces for genetic insights at global scale.

Interactive Report
React

Interactive Report

Explorable sport-genetics report with interactive storytelling.

Uniqgene Team Dashboard
React · MySQL · MongoDB

Uniqgene Team Dashboard

Internal ops dashboard for team and lab coordination.

Uniqgene Mobile App
React Native · Mongo · Node

Uniqgene Mobile App

In-house React Native app for field and lab mobility.

Education 07

Roots in bio and AI.

Master's · GPA 3.62 · 2023–2026

Ege University

Bioinformatics, AI · İzmir, Turkey

Bachelor's · GPA 3.59 · 2017–2022

Ege University

Bioengineering · İzmir, Turkey

Skills 08

A stack that ships.

Each skill with where I used it — not a logo wall.

Agents & AI tooling

  • Claude & CursorDaily agent tooling · SkyStudio, this site · 2024–
  • Spec-driven developmentSpec Kit on feature work · Skymod · 2024–
  • MCP25-app connector catalog · Skymod · 2024–
  • Agent runtimeDurable runs, tool graphs · Skymod · 2024–
  • HITL & evaluationApproval steps, traces · Skymod · 2024–

GenAI & retrieval

  • RAGMilvus, chunk/embed/OCR/rerank · Skymod · 2024–
  • Multi-LLM gatewayOpenAI, Anthropic, Gemini, GOAT · Skymod · 2024–
  • LangChainMulti-provider orchestration · Skymod · 2024–
  • OpenTelemetryTracing across services · Skymod · 2024–

Frontend

  • React & TypeScriptSkyStudio redesign · Skymod · 2024–; Uniqgene · 2022–24
  • React FlowWorkflow editor · Skymod · 2024–
  • React Native & Next.jsMobile and web apps · Uniqgene · 2022–24
  • ViteSkyStudio redesign, this site · 2024–

Backend & APIs

  • Node.js & ExpressSame-origin BFF · Skymod · 2024–; APIs · Uniqgene · 2022–24
  • Python & FastAPIAnalysis agent · Skymod · 2024–; services · Genfoquest · 2023–24
  • SSE streamingResumable agent streams · Skymod · 2024–

Data, cloud & ops

  • PostgreSQL & SupabaseTenant row-level security · Skymod; tuning · Genfoquest
  • MongoDB & MySQLProduct data · Uniqgene · 2022–24
  • RedisModel gateway & RAG stack · Skymod · 2024–
  • AWS S3 & RDSAgent artifacts · Skymod; storage · Uniqgene
  • Azure AD & Google authEnterprise sign-in · Skymod · 2024–
  • DockerContainerized services · Genfoquest · 2023–24

Research 09

Data analysis & AI.

Master's research in medical imaging and clinical statistics — the same rigor I bring to production ML.

  • CaPTk
  • HistomicsTK
  • DICOM / NIfTI
  • NDPI slides

Non-invasive decision support for Glioblastoma Multiforme (GBM) built from magnetic resonance imaging and digital pathology.

A multimodal dataset combines radiomic features from DICOM and NIfTI MR volumes (CaPTk) with patomic morphometrics from NDPI histopathology slides (HistomicsTK). The model predicts 12 critical histopathological parameters from MR alone — with potential to reduce biopsy burden in treatment planning.

Let's build the next thing.

Seeking Senior/Staff AI Engineer roles — full-time, remote or relocation. Also happy to talk GenAI platforms and technical leadership.