Skip to content
View Metaphysicist1's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report Metaphysicist1

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. 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
Metaphysicist1/README.md
Edgar Abasov

Edgar Abasov

I build AI agents and useful software — then make sure they work outside the notebook.

Originally from Georgia. Based in Berlin.
Engineer, co-founder, and lifelong builder.

Portfolio · LinkedIn · X · Email


A short version of the work

My strongest work has one theme: take a noisy, high-friction problem and turn it into a system that can gather evidence, make a decision, and keep going.

That has taken me from chatbots at a major bank, to branded AR products, to autonomous research and developer tools. The stack changes. The standard does not: a useful result should be observable, reproducible, and usable.

Three things worth opening

01 — Paperclaw

The question: how much of the solar capacity promoted around Berlin actually exists?

What I made: an always-on agent that combines the official registry with Sentinel-2 satellite imagery, tracks 571 registered parks, and separates verified generation assets from projects that exist only on paper.

Why it was technically interesting: the difficult part was not one model call. It was keeping a changing source of evidence coherent: retrieval, geospatial checks, ranking, alerting, and a public presentation of what the system believes.

Explore the live monitor · Read the repository


02 — Relay

The question: what happens when a software task is divided before one coding agent becomes the bottleneck?

What I made: a parallel agent workflow that takes one prompt and runs work across isolated git worktrees, giving separate agents their own branch and preventing changes from being mixed together too early.

What I learned: parallelism is only useful with boundaries. Execution, review, merge decisions, and access to the repository need to be designed separately.

View the repository


03 — UniMate

The question: could a confusing admissions process become something a student can actually navigate?

What I made: an AI companion for German university admissions, focused on turning scattered requirements into clear next steps.

Live proof: unimate.study · Source


Selected work

| AI radar | Daily AI briefings and urgent Telegram alerts for an engineering workflow. | | Supply-AI-Agent | Gemini and LangGraph supply-chain agent deployed on Google Cloud. | | Telegram delivery | Encrypted, deterministic delivery with durable duplicate suppression. | | Wortweg | A free daily German word puzzle for A1–C1, with no signup or ads. | | PneuNet | A Cloud Run API for pneumonia detection from chest X-rays. | | ClipVault | A local-first GTK4 clipboard history for GNOME. |

The path behind the projects

  • AI Engineer at Bank of Georgia — raised chatbot accuracy by 15% and shipped Georgian/English chatbot features.
  • Co-founder & COO at stamo.AI — now building branded AR effects for TikTok, Instagram, and Snapchat.
  • Computer science — B.Sc. in Batumi, followed by Erasmus+ Data Science studies at HAW Kiel.
  • Teaching — worked with 80+ children aged 12–14, helping them build games and learn Python.
  • Certifications — AWS Certified Machine Learning – Specialty and Azure Data Scientist Associate.

What I use

Python · LangGraph · FastAPI · Gemini · PyTorch · Docker · Google Cloud · Linux · Git

See more of the shipped work

Neuro Timer · Berlin Flat Search · Clustar · CoverBot


Building something ambitious, useful, or wonderfully difficult?
Let’s talk.

Pinned Loading

  1. Credit_Score_Classifier Credit_Score_Classifier Public

    Web App where you can enter customers parameters and get credit scoring result

    Jupyter Notebook 1

  2. Customer_Churn_Classification Customer_Churn_Classification Public

    Predicting Customer Churn in Telecommunication Industry

    Python

  3. Diamond-Price-Prediction Diamond-Price-Prediction Public

    Flask Web application that predicts diamond price

    Jupyter Notebook

  4. CoverBot CoverBot Public

    RAG based chat interface for easy insight Retriviation

    Python

  5. Fine-tuning-NER-for-En Fine-tuning-NER-for-En Public

    Fine-tuning NER using PyTorch

    Jupyter Notebook

  6. Food-Delivery-Analysis Food-Delivery-Analysis Public

    This repository contains a Power BI report that provides insights into food delivery trends