Skip to content
Available for hire
Christos Prapas
AI / ML
2026

Athens · Software & Applied AI

Software, AIengineer.

Applied ML in production.

I build systems that work in production — and propose the tools teams don't know they need yet. Deep-learning classification, computer vision, agentic RAG, ERP systems.

PythonPyTorchComputer VisionConvLSTMConv3DSentinel-2LangGraphChromaDBQLoRADjangoNext.jsTypeScriptFly MachinesPostgreSQLPlaywrightPythonPyTorchComputer VisionConvLSTMConv3DSentinel-2LangGraphChromaDBQLoRADjangoNext.jsTypeScriptFly MachinesPostgreSQLPlaywright

── At a glance

5+

Years building

production systems

94.2%

Crop accuracy

Sentinel-2 ConvLSTM

10k+

Farmer clients

supported via ERP

4

In production

ERP · CV · RAG · LLM pretraining

Act I · The Context

The ERP

A full-stack ERP for the Greek agricultural sector — supporting 10k+ farmer clients with declarations, subsidies, land registry, and compliance reporting.

Client profile — full farmer record with every detail on one screen.
Client profile — full farmer record with every detail on one screen.
Compliance checks and program scoring across farmer declarations.
Compliance checks and program scoring across farmer declarations.
Ecoscheme eligibility — automated rule checks before submission.
Ecoscheme eligibility — automated rule checks before submission.

Act II · My Proposal

Crop classification
from satellites.

MSc thesis: ConvLSTM + Conv3D models on Sentinel-2 multispectral time-series to classify crops — and automatically flag declarations that don't match the satellite.

The proposal: The ERP handles declarations, but can't verify them. I proposed using satellite imagery — the same data OPEKEPE uses for aerial checks — to classify crop types automatically and surface inconsistencies before they become penalties.

94.2%

Overall Accuracy

on held-out test set

0.96

F1 — Wheat

most common crop class

0.91

F1 — Cotton

highest-subsidy crop

12

Timesteps

Sentinel-2 patches per season

Models trained on the PASTIS benchmark (France) and validated on Greek region patches. Architecture: ConvLSTM encoder + 3D convolutional decoder, pixel-level classification.

Act III · My Proposal

Advisory RAG
for the team.

A LangGraph agentic RAG system that answers Greek agricultural regulation questions with citations — built so advisors don't have to memorize 200-page policy documents.

The proposal: Advisors spent hours per week hunting through OPEKEPE bulletins and CAP regulation updates. I proposed a retrieval system that indexes the authoritative sources and answers questions with source citations — reducing a 20-minute search to a 10-second query.

Demo query →

Τι είναι το ΟΣΔΕ και ποιον σκοπό εξυπηρετεί;

Το ΟΣΔΕ (Ολοκληρωμένο Σύστημα Διαχείρισης και Ελέγχου) είναι το πληροφοριακό σύστημα μέσω του οποίου ο ΟΠΕΚΕΠΕ παραλαμβάνει την Ενιαία Αίτηση Ενίσχυσης (ΕΑΕ), διενεργεί ελέγχους επιλεξιμότητας στα αγροτεμάχια και καταβάλλει τις άμεσες ενισχύσεις της ΚΑΠ προς τους δικαιούχους αγρότες.

ΟΠΕΚΕΠΕ — Οδηγός ΟΣΔΕΚΑΠ 2023–2027 — Άμεσες Ενισχύσεις
Launch Live Demo →

Requires email verification · 20 min session · 10 LLM calls included

Act V · Who I Am

Let's build something.

Christos Prapas

Christos Prapas

Full-stack engineer with applied ML in production. The two acts you just saw are why.

Based in Greece. Open to AI/ML engineering roles in Germany and remotely. Interested in LLM engineering, applied ML, and AI-first product work.

ML / AI

PyTorchComputer VisionConvLSTMConv3DLangGraphChromaDBQLoRAOllama

Backend

PythonDjangoNode.jsFastAPIPostgreSQLSQLite

Frontend

Next.jsReactTypeScriptTailwind CSSGSAP

Infra

DockerFly.ioVercelPlaywrightGitHub Actions

Built with Next.js + GSAP

christosprapas.gr

Christos Prapas — Full-stack Engineer & Applied ML