Erol Sen

AI Engineer, Izmir, Turkey

i build agentic ai systems: multi-agent orchestration on langgraph and crewai, graph rag over neo4j, and memory engineering.

Get in touchRead the writing

Focus

What I work on

01

Agentic systems

Multi-agent orchestration on LangGraph and CrewAI: routers, supervisors, repair loops, and structured memory for long-running workflows.

  • LangGraph
  • CrewAI
  • Multi-agent
  • Memory
02

Conversational AI

Production chat assistants: streaming responses, tool calls, and retrieval-grounded answers with guardrails and evaluation.

  • Streaming
  • Tool use
  • RAG
  • Guardrails
03

Retrieval & knowledge graphs

RAG and Graph RAG over Neo4j, with deterministic rules and LLM-as-Judge evaluation for correctness-critical answers.

  • Neo4j
  • Graph RAG
  • LLM-as-Judge
04

Edge computer vision

Low-latency vision on NVIDIA Jetson: YOLO, OpenCV, and VLM pipelines for real-time inference at the edge.

  • YOLO
  • TensorRT
  • Jetson
  • VLM

Notes

Writing

All writing
14 min readTransformersAttention

Attention

The word "bit" gets the same vector in "the dog bit" and "a little bit", so attention rewrites each token by letting it look at its neighbors. Query, Key, Value, scaled dot-product, causal masking, and 16 parallel heads, step by step. An animated explainer.

9 min readTransformersAttention

Positional Encoding

Attention reads a sentence as a bag of words, so "the dog bit the man" and "the man bit the dog" look identical to it. How a transformer learns word order, from additive embeddings to RoPE. An animated explainer.

8 min readNVIDIAComputer Vision

LocateAnything

NVIDIA's LocateAnything draws a whole object box in one stroke instead of one coordinate at a time. An animated, hands-on explainer.

16 min readAgentsMemory

How AI agents remember

A field guide to agent memory: the kinds of memory an agent keeps, where each one lives in Postgres, when the agent reads or writes it during a single turn, and when one Postgres stops being enough.