Portrait of Kosmas Meta

Hey, I'm Kosmas.Senior AI Engineer at saas.group

I ship AI agents to production and make them measurable, with evals on real traffic and guardrails that keep them honest. Before that, I wrote C firmware for IoT devices.

Contact me here Resume

About me

I'm Kosmas (Louis-Kosmas Meta on paper), a 27-year-old Senior AI Engineer at saas.group. I studied Electrical and Computer Engineering at the Aristotle University of Thessaloniki, finished my M.Eng. with Great Distinction (top 6%), and spent a semester on Erasmus in Madrid.

I started out writing embedded firmware in C for IoT devices. It taught me to care about things most ML engineers skip: memory limits, failure modes, and systems that have to work when nobody is watching. I kept that habit when I moved into AI.

At saas.group I lead and build AI features across several of our brands: AI agents, RAG pipelines, and NLP features that ship inside real products. I own them from the first prototype to production monitoring. I care a lot about evaluation and observability, because shipping an LLM feature without them is just hoping it works.

Outside work I play football, play video games, and lately I've been learning video editing. The next travel vlog is taking a while, so until then, grab some popcorn and watch my latest one from Iceland on my YouTube channel.

I built this site to share what I'm working on and meet people. Say hi on LinkedIn or by email.

My projects

Txtify

Open-source app that turns audio and video into text with Whisper. It handles multiple languages and output formats, translates with DeepL, and you can self-host it. 125+ stars on GitHub.

  • Python
  • Whisper
  • Hugging Face
  • DeepL
  • FastAPI
View project (opens in a new tab)
Txtify transcription app interface showing audio-to-text conversion

JoyFill

Open-source web app that suggests positive ways to finish a sentence. BERT fills in the blank, DistilBERT checks that the result actually sounds positive.

  • Python
  • BERT
  • FastAPI
  • Docker
  • Hugging Face
View project (opens in a new tab)
JoyFill web app suggesting positive sentence rewrites with sentiment analysis

Neural Networks - Deep Learning

University project on digit recognition (0 to 9), comparing SVM, kNN, NCC, and neural networks with PCA.

  • Python
  • NN
  • Deep Learning
  • Digit Recognition
View project (opens in a new tab)
Neural network digit recognition results comparing SVM, kNN, and sequential models

Non Local Means

Parallel programming exercise: the Non-Local Means image denoising algorithm in CUDA, with up to a 200x speedup.

  • C
  • CUDA
  • Image Processing
View project (opens in a new tab)
Non-Local Means image denoising comparison showing CUDA-accelerated results

My skills

LLMs & AI

  • LLMs
  • AI agents
  • RAG
  • Evals
  • LLM-as-a-judge
  • NLP
  • LangChain
  • LangGraph
  • LangSmith
  • Langfuse
  • AWS Bedrock

Machine learning

  • PyTorch
  • TensorFlow
  • HuggingFace
  • sklearn
  • NumPy
  • Pandas

Backend & data

  • FastAPI
  • Celery
  • Redis
  • SQL
  • PostgreSQL
  • pgvector
  • MySQL
  • MongoDB

Infra & DevOps

  • AWS
  • Azure
  • Docker
  • Kubernetes
  • Prometheus
  • Grafana
  • Git
  • Bash

Languages

  • Python
  • TypeScript
  • JavaScript
  • C
  • C++
  • LaTeX

My experience

  1. Senior AI Engineer - saas.group

    Remote, Anywhere

    • Lead and contribute to AI projects across several SaaS brands and central teams, owning them end to end: agents, RAG pipelines, and NLP features, from first prototype to production monitoring.
    • One example is Pollen Create for Beekast, where I work directly with the brand's engineers and product team.
    • Work with Python, FastAPI, LangChain, LangGraph, LangSmith, PostgreSQL/pgvector, Redis, Celery, and AWS Bedrock and OpenAI models, plus TypeScript on the product side.
    • Use Langfuse and LangSmith tracing with evaluation suites to gate releases and catch regressions before users do.

  2. AI Engineer - Cogninn

    Larissa, Greece

    • Led development of AutoDL, a platform that takes images, text, video, and tabular data and automates the ML/DL pipeline around them, improving training accuracy and reliability.
    • Implemented classification, object detection, segmentation, OCR, and text analysis, then added video object detection and tabular regression, which raised performance by 15%.
    • Built the REST APIs with FastAPI, backed by MongoDB and SQLite.

  3. Freelancer

    Remote, Greece

    • Delivered deep learning projects for clients across NLP, computer vision, and data analysis.
    • Taught more than 50 university students AI, maths, and programming, with 95% positive feedback.

  4. Embedded Systems Engineer - NET2GRID

    Thessaloniki, Greece

    • Worked in a 6-person team building an embedded firmware library in C for ESP32 IoT devices.

  5. Embedded Systems Engineer Intern - NET2GRID

    Thessaloniki, Greece

    • Set up a Greengrass Group for IoT devices and connected it to AWS (IoT Core, DynamoDB, API Gateway, Lambda).
    • Built MQTT bridge connections and added a Keras-based predictive model on top of the device data.

Contact me

Please contact me directly at louiskmeta@gmail.com or through this form.