Skip to content
View SpiliosDimakopoulos's full-sized avatar

Block or report SpiliosDimakopoulos

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. Markdown is supported. 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

Spilios Dimakopoulos

I build AI products end-to-end — from requirements to live deployment.

Products

  • FinTwin — digital twin of 390+ S&P 500 companies' finances from SEC filings, 5,000 crisis simulations live in the browser, rebuilt weekly for 0€ · Live · Code
  • AI layer for a children's teddy bear that speaks, listens and tutors kids — entirely offline
  • Live AI university advisor — Llama-3.3-70B, Docker, 0€/month · Live
  • Skill-swap platform — 6 user interviews before writing a line of code · Live
  • HR portal for 2,400 security guards — live, costs nothing to run · Live
  • Conversational analytics platform — upload a dataset, ask in plain language, get instant analysis · Live
  • IoT stress detection system — smart ring adjusts lighting and music automatically, no manual input · Live

Growth

  • Top 0.5% on Adobe Stock out of 342,000 creators — automated pipeline -> 200k assets live, zero ads · Profile
  • 245M+ organic impressions across 6 platforms — zero budget, zero ad spend

Engineering

  • Upstreamed the TwinLens fix to circuit-tracer, the attribution-graph library · PR
  • 70 merged PRs across 13 public repos — Mainly LangChain, DSPy (36.4k★), Dify (144k★) · GitHub
  • 400+ LeetCode problems solved - daily problem solving routine :) · Profile

Research

  • TwinLens — showed LLM attribution graphs mispredict interventions because they drop two derivatives; restoring them lifts accuracy ρ 0.20 → 0.53 on GPT-2, replicated on Qwen3 · Paper · Live · Code
  • Showed that balance-sheet stress models mostly measure persistence — and where a simulation twin still beats simple cash-flow rules, every year 2014–2023 (2,661 firm-years) · Paper
  • Proved a locally-correct AI can be systematically unfair over time — EU AI Act aligned · Paper
  • Validated new startup efficiency metric on 5.6M UK company records · Paper
  • Peer Reviewer: Journal of the International Council for Small Business
  • Multiple manuscripts currently under peer review.

  • Currently fine-tuning the world's most accurate Ancient Greek LLM, trained on the entire classical corpus
  • Plus architecting autonomous AI pipelines for end-to-end digital product deployment.

BSc Management Science & Technology — AUEB, 2026 Open to: AI x Product roles Portfolio Linkedin

Pinned Loading

  1. greekbook greekbook Public

    Markdown → beautifully typeset Greek books and theses (PDF, PDF/A, EPUB) with Typst. Greek hyphenation, «“quotes”», accentless caps, citations, live editor, in-browser demo.

    Python