Skip to content
View KNHNF's full-sized avatar
⚡
Open to sponsored graduate roles
⚡
Open to sponsored graduate roles

Block or report KNHNF

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
KNHNF/README.md

Karan Homayounfar

Typing headline

MSc Data Science at UWE Bristol, working on GB electricity market forecasting with Python, C++ and SQL. BSc Computer Engineering, 1st in class (GPA 18.87/20).

Portfolio: knhnf.com

Energy and trading systems

  • MSc dissertation (repository private until assessed): forecasting the official NESO daily GB balancing cost one day ahead, using only data published before the delivery day. A Ridge model is tested against simple baselines with a fixed-b Diebold-Mariano test, and every forecast is saved with a timestamp before delivery. Due January 2027.
  • GB energy forecasting reproductions: rebuilds three published GB balancing-market forecasting studies on real data. Only the Lucas MAE on its original window matched the paper; the other reproduced results were worse.
  • GB BM data: typed Python clients for Elexon BMRS, NESO and Carbon Intensity data. It raises an error on the three BMRS endpoints that ignore historical date parameters, instead of returning misdated data.
  • C++ order book simulator: price-time-priority matching, cancellation, invariant tests and seeded workloads. A single-threaded local benchmark measured median throughput from 3.36 to 5.45 million synthetic events per second.

Data and machine learning

  • England crime and deprivation analysis: Spark pipeline over 17.8 million police records. It found a 4.8 times crime-rate gap across deprivation deciles and identified Greater Manchester as missing data rather than a low-crime result.
  • EPC retrofit potential: compares four classifiers under nested cross-validation and a temporal holdout. Random Forest reached 0.9721 ROC-AUC and 0.8423 macro F1 on the held-out period.

Tools

Python, C++, SQL, Apache Spark, scikit-learn, XGBoost, pandas, FastAPI, MongoDB and GitHub Actions.

LinkedIn | Email | knhnf.com

Pinned Loading

  1. gb-bm-data gb-bm-data Public

    Typed Python clients for Elexon BMRS, NESO and Carbon Intensity. Live-only endpoints fail loudly.

    Python

  2. gb-energy-forecasting-reproductions gb-energy-forecasting-reproductions Public

    Rebuilds three GB balancing-market forecasting papers on real BMRS data. Only one matched.

    Python

  3. cpp-order-book-simulator cpp-order-book-simulator Public

    C++20 limit order book with price-time priority, partial fills, cancellation, benchmarks and CI.

    C++

  4. epc-retrofit-potential-ml epc-retrofit-potential-ml Public

    Predicting UK residential retrofit potential from EPC open data using Random Forest, XGBoost, Logistic Regression and SVM under nested cross-validation

    Jupyter Notebook 1

  5. england-crime-deprivation-analysis england-crime-deprivation-analysis Public

    Spark pipeline over 17.8M police records linked to deprivation. Finds a 4.8x crime-rate gap by decile.

    HTML

  6. sme-reception-llm sme-reception-llm Public

    LLM pipeline for SME voice assistant - QLoRA fine-tuning, FastAPI backend, Faster-Whisper STT

    Python