I'm an AI Engineer at GET, specializing in converting raw data and large language models into production-grade, observable AI systems. My work sits at the intersection of NLP, RAG architectures, and agentic workflow orchestration — with a rigorous MLOps mindset across GCP, Azure, and AWS.
With a background spanning civil engineering → MBA → data science → AI engineering, I bridge deep technical implementation with measurable business outcomes.
What I actually ship:
- 🔍 RAG Pipelines — hybrid retrieval (dense + SPLADE++), reranking, agentic query planning
- 🤖 Agentic Systems — multi-agent orchestration with LangGraph, CrewAI, Google ADK
- 📊 LLMOps Platforms — token optimization (60–90% cost reduction), caching, drift monitoring
- 🚗 Automotive AI — vehicle log analysis, real-time sensor data processing, production dashboards
| 🎯 Domain | 📈 Achievement |
|---|---|
| LLM Cost Optimization | 60–90% token cost reduction via KV-cache + batch routing |
| RAG Retrieval | SPLADE++ learned sparse retrieval + hybrid reranking in production |
| Agentic Workflows | Multi-agent orchestration across LangGraph, CrewAI, Google ADK |
| ML Pipeline Scale | End-to-end MLOps on GCP · Azure · AWS with full observability |
| Automotive AI | Real-time vehicle log analysis + predictive maintenance dashboards |
| Certifications | IBM AI Engineering · Certified Data Scientist & ML Engineer |
| Project | Stack | Description |
|---|---|---|
| 🔍 Production RAG Platform | LangGraph · Qdrant · SPLADE++ · FastAPI | Agentic query planning + hybrid retrieval + reranking pipeline for research document analysis |
| 🤖 LLMOps Token Optimizer | LangChain · LMCache · Redis · Prometheus | 60–90% cost reduction framework with semantic caching, KV-cache, and batch routing |
| 🚗 Automotive Log Intelligence | PostgreSQL · pgvector · Neo4j · React/TS | Vehicle log analysis combining vector similarity + graph relationships for root cause detection |
| 🧠 Multi-Agent Orchestrator | LangGraph · CrewAI · Google ADK · FastAPI | Supervisor-subagent architecture for parallel, autonomous workflow execution |
| 📡 On-Call Alert Intelligence | LangGraph · PagerDuty · Grafana · Slack | LLM-driven alert routing and triage system with contextual escalation logic |
| ⏳ Time-Series Forecasting | Prophet · LSTM · NeuralProphet · MLflow | Multi-horizon forecasting for automotive sensor data with drift detection and alerting |
📂 Full portfolio → datascienceportfol.io/kostas696
📌 2025–2026 Engineering Priorities
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔬 Learned Sparse Retrieval (SPLADE++) ──── Advanced RAG Retrieval
🧠 BDI Mental State Modeling ──── Cognitive Agent Architectures
💰 Context Optimization & KV-Cache ──── LLM Cost Engineering
🔄 Agentic Workflow Orchestration ──── Plan-First Multi-Agent Systems
🏗️ Google ADK Production Deployments ──── Agent Infrastructure at Scale
📊 LLM Evaluation Frameworks ──── LLM-as-Judge + RAGAS
"Shipping AI systems that are observable, reproducible, and measurably useful."