Public work

Talks, open-source software, datasets, and demonstrations.

Talks & invited lectures

A short list of recent talks on production GenAI, retrieval, and source-code intelligence.
  • 2025
    Invited talk — Queen Mary University of London (DERI)
    CodeQUEST: Iterative Evaluation and Enhancement of Code Quality with LLMs. Hosted by Prof. Greg Slabaugh (Director, Digital Environment Research Institute). Talk covered the dual Evaluator / Optimizer framework behind CodeQUEST, deployment lessons from running LLM-based code review at scale, and open research questions on calibrated code-quality scoring.

Open source & datasets

Publicly released code and datasets from across the work.
  • claude-academic-prose — A Claude Code skill that enforces clear, evaluable scientific writing: a plain-empirical voice profile, rules that strip hype and rhetorical flourish, and structural rules that make every claim traceable to its evidence.
  • BitBudget Benchmark — An open embedding-compression benchmark extending my PhD learning-to-hash line; single-bit codes with re-ranking stay lossless at 32× compression.
  • The Corpus is the Model — Interactive image annotation demo based on my PhD research.HF Spaces
  • arxiv-acronym-gen — 10K+ acronym/expansion pairs for training LLMs.HF Datasets
  • arxiv-tagged-papers — Semantic keyword tags for 3M+ arXiv papers.HF Datasets
  • SpamT5 — LLM-powered email spam detection (FinLLM @ IJCAI 2023).
  • CodeQUEST — LLM-powered code-quality evaluation/improvement (ISSREW 2025).
  • ai-prepline — Text preprocessing pipelines for downstream NLP.
  • DeepLPF — CVPR 2020 image-enhancement codebase.
  • CURL — ICPR 2020 neural curve layers.
  • SIDGAN — ECCV 2020 low-light video enhancement.
  • SKLCRM — Sparse kernel relevance model (Best Paper, ICMR 2014).
  • CryptoPanda — Cryptocurrency scanner backtest.
  • Satire Classifier — Naive Bayes NLP demo.