Log
A working ledger of what I am building, writing, and wrestling with.
August 2026 UC activity manuscript, circulated
Completed the Bayesian multilevel manuscript and sent it to co-authors for a final review before submission.
August 2026 Next motility question, taking shape
Moved the next UC motility project into concept development: whether MRI motility helps explain symptoms, and how it relates to inflammation, structural disease, endoscopy, and histology.
August 2026 Review directions, scoped
Started mapping critical reviews around persistent mural disease despite mucosal healing, how MRE activity indices are measured and validated, the reference standards behind them, and inflammation–fibrosis separation.
July 2026 Loop engineering?
Looked into where a verification loop actually belongs in the lineage pipeline. Landed on Stage 6 selection as the right place for it, and just as clearly on writing as the wrong one.
July 2026 Content format taxonomy, drafted
Worked out a full backlog of content formats the tooling could support: lineage pieces, pipeline meta-posts, living reference pages, methods close reads, a reproducibility ledger. Filed it as backlog, not a prerequisite. The first piece still comes first.
June 2026 Lineage module, stages 3 through 6 built
Reframed the pruner after similarity-based cuts failed to separate on-thread from off-thread papers. The distinction turned out to be structural and temporal, not topical. Stage 6 shipped as a two-step selection-plus-narration system with a strict anti-hallucination contract.
June 2026 Corpus persistence, diagnosed
Untracking papers.db to stop binary merge conflicts quietly broke corpus accumulation along with it. Fixed the conflict, lost the persistence, now rebuilding the data branch and reseeding from Zotero.
Juune 2026 Patient-level triage, ruled out
Spent real time on whether patient-level accuracy claims could be supported, testing worst-segment, mean, and count-based rollups against each other. They agreed on outcomes barely 6 times in 10, which settled it: no independent patient-level ground truth means no patient-level accuracy metric.
June 2026 Risk Stratifier app, v0.5.0
Replaced two inconsistent hardcoded parameter files with a single artifact-driven design, so the app now loads directly from the notebook’s exported JSON. Pushed a UI cleanup alongside it and cut a new Zenodo release.
June 2026 Methods and Results, drafted
Worked through the manuscript’s Methods and Results paragraph by paragraph, locking prior specification, sampling settings, and exclusions along the way. Delivered as a validated file with placeholders marked for what’s still missing.
June 2026 ESGAR 2026, presented
Dr. Bhatnagar presented the work in Montpellier on June 11th and it landed well.
June 2026 ibd-reports, section-splitter bug fixed
Found a quiet parsing bug where soft line breaks were hiding section headers on later lines. Fixed it and re-ran the affected phases; recovered dozens of impressions that had gone missing.
June 2026 ibd-reports, duplicate review closed out
Manually reviewed thousands of fuzzy-matched filename clusters to confirm genuine multi-report patients versus false merges. Tedious, but it locks the input for later longitudinal linkage.
June 2026 ibd-reports, de-identification validated
Passed the second manual validation layer on the de-identification scrub with a clean stratified read. Locked the design for the final anonymous-ID rekey.
May 2026 Activity project, manuscript start
Started structuring the manuscript. The first pass is always the longest.
May 2026 ibd-digest, shipped
Automated literature workflow on GitHub Actions, semantic embeddings to surface newly published IBD imaging papers each day. Letting it run for a few days before deciding the next steps.
May 2026 ESGAR 2026
Closed out slides for two first-author orals at ESGAR 2026 in Montpellier and sent to co-authors.
May 2026 Motility manuscript, submission queued
The manuscript reached a submission-ready draft. We decided to circulate it for one final read by all authors after ESGAR 2026. The delay bothers me, but the paper will be better for it.
May 2026 Edge cases on a clinical extraction pipeline
Spent a few days handling legacy formatting in historical reports that breaks naive parsing. Most of the work in agentic systems lives here, not in the model calls.
April 2026 Motility manuscript, final pass
Last structural revisions and cross-site checks before submission. The hardest part has been trimming a well-written paragraph.
March 2026 Personal site, live
Quarto build, custom theme, treated as both a professional presence and a digital garden. The harder decision was structural, not technical: deciding what counts as ready to publish.
Last entry: August 2026. Updated as work ships or stalls.