Skip to content
View coderback's full-sized avatar

Highlights

  • Pro

Block or report coderback

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
coderback/README.md

Timothy Oluwatobi Ojebiyi

CS Graduate · Software Engineer · Aspiring ML Researcher

Open to ML engineer, Software engineer and Quant developer roles

Portfolio · LinkedIn · ORCID · Email


Computer Science graduate from UWE Bristol (AI/ML specialisation) and founder of Rootlayer. I build and ship ML systems, implement papers, test their claims, and decide how a result will be judged before I see it: pre-registered hypotheses, leakage checks, and baselines a model has to beat.

Research

News Timing Cannot Explain It: A Pre-Registered Replication of Zhang (2025)
Working paper, 2026 · DOI 10.5281/zenodo.24087562 · code + paper
A published FinBERT + XGBoost strategy reported Sharpe ratios of 4.6–5.9. In a pre-registered replication, none of 18 variants beats random-sign trading after costs.

Fair on Paper: What a Prop-Firm Challenge Is Worth, and Who Pays for It
Working paper, 2026 · DOI 10.5281/zenodo.23374897 · reproduction package
Prices a published contract on 20 years of intraday data, with 24 pre-registered hypotheses and the buyer's optimal policy found by backward induction.

Built

  • Rootlayer: founder of an agentic web-compliance SaaS that audits sites for WCAG 2.2 AA accessibility and GDPR/EAA privacy
  • Tubesense (live demo · repo): places Tube trains to a median of ~2 m from phone sensors alone, then an agent drafts a cited evidence pack that a verifier checks. Encode London Hackathon 2026
  • Thespis (live demo · repo): NPCs with memory and beliefs that can be false, for game developers. Cambridge × Arcade AI Hackathon 2026
  • Unsaid (repo): semantic detection of risk disclosures removed from 10-K filings. It flagged hedging language Silicon Valley Bank dropped 14 days before it collapsed

Currently

  • GBE: a graph embedding engine for financial-crime and risk modelling (GNNs, leakage-tested pipelines)
  • Member of the Cohere Labs Open Science Community

Interests

Graph neural networks · ML interpretability · generative models · agent memory · evaluating ML in finance honestly

Tools

ML: Python · PyTorch · PyTorch Geometric · scikit-learn · XGBoost · Hugging Face · LaTeX
Engineering: TypeScript · React / Next.js · FastAPI · Django · PostgreSQL · Docker · C++

Pinned Loading

  1. ControlNet-Implementation ControlNet-Implementation Public

    From-scratch, parity-verified ControlNet (Stable Diffusion v1.5) with end-to-end training, inference, and evaluation on COCO Canny conditioning.

    Jupyter Notebook

  2. fifa-wc2026-prediction-model fifa-wc2026-prediction-model Public

    Probabilistic 2026 FIFA World Cup forecasting engine using Elo + Dixon–Coles goals modeling, gradient-boosted ensemble blending, and 50k-run Monte Carlo simulation with rigorous walk-forward/backte…

    Jupyter Notebook

  3. Unsaid Unsaid Public

    AI 10-K disclosure diffing: compares Item 1A/7A across years and surfaces REMOVED, SOFTENED, and NEW risk language using semantic alignment + LLM judgment.

    TypeScript

  4. prop-challenge-pricing prop-challenge-pricing Public

    Reproduction package for 'Fair on Paper: What a Prop-Firm Challenge Is Worth, and Who Pays for It' (Ojebiyi, 2026)

    Python

  5. Tubesense Tubesense Public

    AI agent that turns ordinary Tube rides recorded on a phone into a privacy-safe, section-by-section noise map of the London Underground, locating the train without GPS. Gemini 4 Argon checks TfL's …

    Python

  6. zhang2025-replication zhang2025-replication Public

    Pre-registered replication of Zhang (2025), arXiv 2505.16136: FinBERT + GDELT + XGBoost macro alpha (not replicated)

    Python