Skip to content
View erdincka's full-sized avatar
  • HPE
  • London, UK
  • 18:29 (UTC +01:00)
  • X @erdincka

Block or report erdincka

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
erdincka/README.md

Erdinc Kaya

Principal Solutions Architect working on enterprise AI infrastructure — GPU platforms, agentic systems, and the unglamorous problem of making them run safely inside real organisations.

Most of what I build starts as a question from a customer: can this actually work here, and can we prove what it did? The repos below are public-safe versions of that work.


What I'm building now

Enterprise agent observability — instrumenting multi-agent workflows with OpenTelemetry's GenAI semantic conventions on a self-hosted stack (OpenLIT, ClickHouse, LiteLLM, Perses). The question I'm chasing: how do you produce audit-grade agent traces when you're not permitted to record the content? The specification leaves that tradeoff to implementers. Enterprises with real audit requirements don't have that luxury.

Work in progress — notes and findings published as I go.


Selected work

meetings Multi-agent AI system where each agent runs in its own gVisor Kubernetes sandbox. LangGraph orchestration, RAG with pgvector, FastAPI + Next.js.
defence-ops Tactical operations dashboard — four live video feeds analysed by a vision-language model. KServe inference endpoints, Kafka telemetry, microservice architecture.
satellite Headquarters and a disconnected edge sharing only what matters — broadcast descriptions, request the data. Edge-to-core streaming with VLM integration.
manufacturing-co Factory telemetry lakehouse with an isolated AI assistant. Medallion architecture on Kafka, Iceberg, S3.
catchx Fraud detection pipeline — medallion architecture across Kafka, HBase, Iceberg, Hive and Delta Lake.
lawfirm-co Case management with AI-assisted document analysis. Next.js, FastAPI, PostgreSQL.

Most of these are built for a specific enterprise AI platform and say so in their README — they're reference implementations rather than turnkey projects. The patterns transfer; the deployment assumes that platform.


Writing

I publish build notes and architecture write-ups at erdincka.github.io.


Background

20+ years across infrastructure, storage, data platforms and AI, working with organisations in nearly every sector. Currently Principal Solutions Architect, Data & AI. NVIDIA-Certified Associate (AI Infrastructure and Operations). Spoke on agentic AI at AI Summit London 2025.

Python · Go · TypeScript · Kubernetes · OpenTelemetry · LangGraph · MCP

📍 London

LinkedIn

Pinned Loading

  1. meetings meetings Public

    Multi-agent meetings where each agent runs in its own gVisor Kubernetes sandbox, and the cluster — not the prompt — decides what it may do

    Python

  2. defence-ops defence-ops Public

    A tactical operations dashboard: four live video feeds analysed by a vision-language model, on infrastructure you control

    TypeScript

  3. lawfirm-co lawfirm-co Public

    Case management with an AI assistant that reads the firm's own case files, for settings where documents cannot leave

    TypeScript 1

  4. catchx catchx Public

    A fraud-detection pipeline on HPE Data Fabric: streams and CSVs through a bronze/silver/gold medallion architecture

    TypeScript

  5. manufacturing-co manufacturing-co Public

    A medallion lakehouse for factory telemetry on HPE Data Fabric, with the AI assistant kept inside the data boundary

    TypeScript 1

  6. satellite satellite Public

    Headquarters and a disconnected edge sharing what matters — broadcast descriptions, request the data. Runs in one container

    Python