Weeknotes 2026 W14

Week range: 2026-03-31 to 2026-04-06 (draft window)

Headline

This week was less about launching shiny pages and more about making the site behave like a living system. I kept the output honest: timeline first, long posts only when evidence and story quality were both strong.

Entries

1. HPC thread stayed evidence-first
   I kept the cache story tied to reproducible artifacts and avoided adding new claims beyond what is already in portfolio/hpc-cache-research.md and the published HPC post.

2. OpsPilot became a stronger long-form candidate
   The strongest angle was not "agent magic." It was contracts, allowlists, and test boundaries. That became a backfill draft candidate.

3. Hackathon and NCAA stayed in motion, not repetition
   Both are already published as full posts. This week they are timeline references, not recycled essays.

4. Backfill got clustered, not spammed
   I grouped related evidence into fewer, stronger stories and rejected tiny duplicate updates.

5. Portfolio archive got expanded paths
   I prepared backfill project files so each major project has a consistent short record and publish target.

6. Blog queue moved to five draft candidates
   I selected five backfill blog drafts that are evidence-backed and useful, then routed lower-confidence items to internal only.

7. Image flow defaults to real assets
   I mapped image prompts only where assets are missing and marked generated images as optional, never fake screenshots.

8. Wix routing now has explicit status logging
   I separated API-writable, manual, and blocked actions with concrete run logs.

9. Desktop repo sync pass surfaced real blockers
   I pushed most connected repos, filtered out blocked large archives and secret-bearing logs, and documented the remaining divergence in the Clap main branch.

Technical

• Built structured run outputs under content_ops/phase3_run_2026_03_31/.
• Added publish-ready weeknote JSON payload for Option A and Option B routing.
• Generated wix_blog_posts/<slug>/ payload folders for backfill posts, including Clap, QERS, and NCAA seed-model follow-ups.

Academic

• No new grade or transcript claims were added.
• Existing academic facts stayed tied to canonical and approved artifacts.

Career and leadership

• No new metric claims were published.
• Resume-backed topics remained in safe language.

Reading and notes

• Reused humanizer constraints from SKILL.md and removed brochure style phrasing.

Wins

• One coherent weekly output and five long-form draft candidates without adding unsupported claims.

Friction

• The easy mistake was turning everything into a blog post. I had to keep saying no and route most movement into timeline format.

Next focus

• Publish one weeknote and at most two backfill blog posts first.
• Attach real images where available before any generated fallback.

Humanizer audit notes

• Remaining AI tells spotted: a few list items sounded too polished and procedural.
• Revision applied: tightened sentence rhythm, used first person in key spots, added concrete tradeoff language, removed generic "impact" phrasing.
• Punctuation check: no em dash characters and no double-hyphen punctuation patterns.

<!-- status: draft week_id: 2026-W14 evidence_tier: moderate source_cluster: Phase 3 pipeline — timeline, backfill, relevancy, publish queue -->

Week of 2026-03-31 (current)

Week range

2026-03-31 through 2026-04-06 (current week draft; adjust end date when closing the week).

Headline summary

I shipped the next layer of the content system: weekly timeline files, a backfill manifest, relevancy routing, blog and image pipelines, and a publish queue that points to Wix-ready paths. The goal is motion you can trust, not a bigger archive.

Technical work

• New ops docs: timeline_schema.md, relevancy_rubric.md, content_router.md, blog_generation_rules.md, image_generation_pipeline.md.
• Registry JSON for publish status; optional stdlib scan helper under scripts/.

Projects progressed

• Website_Per: Timeline + projects archive page drafts; blog index touchpoints; backfill blog drafts in blog_drafts/backfill/ (evidence from portfolio, not new facts).

Research / academic

• No new lab measurements claimed this week; timeline references existing HPC narrative only.

Career / leadership / recruiting

• Nothing new to publish without resume or approval-queue clearance.

Reading / papers / books

• Re-read humanizer checklist (SKILL.md) before saving generated drafts.

Personal development

• Prefer honest friction sections over motivational filler.

Wins

• Six backfill weeks drafted and one current week file; three long-form drafts queued for review before any publish.

Friction

Backfill wants to produce twenty tiny updates; the router caps noise and forces clustering.

Next focus

Run the first real weekly delta scan on allowlisted folders, then one timeline entry from fresh evidence only.

<!-- status: backfill week_id: 2026-W13 evidence_tier: moderate source_cluster: Website_Per content system, publish-ready packs -->

Week of 2026-03-23

Week range

2026-03-23 through 2026-03-29.

Headline summary

The site stopped being only a theme and became a pipeline: canonical profile, allowlisted scans, fact registry, publish-ready blocks, and a checklist that refuses to invent metrics.

Technical work

• Markdown and JSON ops in Website_Per: publish_ready/*, content_ops/*, pages/*, portfolio case files.

Projects progressed

• Personal site: Humanized copy, SEO tables, Wix publish map, and MCP-ready publication notes where API allows.

Research / academic

• No new lab claim this week; content work supported the same evidence story as HPC posts.

Career / leadership / recruiting

• Resume-backed facts only; approval queue for anything not on resume.

Reading / papers / books

• Internal: style and humanizer pass (SKILL.md) applied to drafts.

Personal development

• Writing discipline: story without brochure voice.

Wins

• Three blog drafts aligned to repos and competitions; portfolio entries grounded in case files.

Friction

Separating “interesting to me” from “safe to publish” took longer than editing prose.

Next focus

Add motion over time: weekly timeline and backfill without turning the site into a metrics flex.

<!-- status: backfill week_id: 2026-W12 evidence_tier: moderate source_cluster: Clap_OpsPilot workflow agent -->

Week of 2026-03-16

Week range

2026-03-16 through 2026-03-22.

Headline summary

OpsPilot stayed the slow, honest project: YAML packs, memory, guarded execution, and tests on the pieces that touch the real world. “Agent” is a label; contracts are the hard part.

Technical work

• Python and SQL for durable state; pytest for risky paths; Jetson-class deployment kept as a constraint, not a fantasy datacenter (see portfolio/workflow-agent-opspilot.md).

Projects progressed

• Workflow agent (OpsPilot): Pack format, observe-propose-execute loop, investor-walkable prototype on my machine.

Research / academic

• Parallel thread: HPC measurement honesty informed how I think about logging and allowlists here.

Career / leadership / recruiting

• None specific this week in public-safe facts.

Reading / papers / books

• Internal README and milestone docs in ~/Desktop/Clap_OpsPilot/ (allowlist).

Personal development

• More patience with safety vs usefulness as a dial, not a switch.

Wins

• Clear hardening list next to a working demo: boring on purpose.

Friction

Convenient shortcuts want to become incidents; allowlists and tests only cover what you write down.

Next focus

Ship one more pack with real-world IO and a failure mode table.

<!-- status: backfill week_id: 2026-W11 evidence_tier: strong source_cluster: Claude Hackathon March 2026 -->

Week of 2026-03-09

Week range

2026-03-09 through 2026-03-15.

Headline summary

Forty-eight hours on a telehealth-flavored prototype: knowledge graph in the middle, Twilio where the story needed real comms, Claude for language-heavy flows, and a front end a judge could click without reading JSON. Demo-grade, not a regulated medical product.

Technical work

• Integrated graph visualization, notification flows, and traceable repo paths (see portfolio/telehealth-hackathon-march-2026.md, published retrospective).

Projects progressed

• Claude Hackathon: SDD-style scope doc kept us from shipping a hospital information system in a weekend.

Research / academic

• Hackathon only; no course credit claimed here.

Career / leadership / recruiting

• Presentation to judges; no public claim of FDA clearance.

Reading / papers / books

• Repo docs and SDD; internal teammate details omitted from public site.

Personal development

• Learning to show empty graph regions as empty, not as fake depth.

Wins

• Coherent demo arc; boxes we could open on stage even when embarrassing inside.

Friction

Time, type errors, and the gap between “cool graph” and “clinically useful graph.”

Next focus

Reuse the scope discipline on longer-horizon builds (OpsPilot, research).

<!-- status: backfill week_id: 2026-W07 evidence_tier: strong source_cluster: NCAA Final Four Analytics Challenge -->

Week of 2026-02-09

Week range

2026-02-09 through 2026-02-15.

Headline summary

I entered the NCAA Final Four Analytics Challenge because leaderboards are rude in a useful way: a deadline, a metric, and a public row with your name next to a number that will not love you back.

Technical work

• Python models on Zerve; slides and a presentation transcript so reasoning survived the event (see portfolio/ncaa-final-four-analytics-2026.md, blog draft on overfitting).

Projects progressed

• NCAA analytics: Public leaderboard submissions; feature-framing to avoid leakage into “future information.”

Research / academic

• Framed as competition work, not a course project.

Career / leadership / recruiting

• Networking at the event is internal detail unless I have clearance to name specifics.

Reading / papers / books

• Competition rules and data docs; no extra reading list for public timeline.

Personal development

• Humility practice: March eliminates stories that only worked last year.

Wins

• Submitted, presented, documented failures alongside wins in the deck (per blog themes).

Friction

Single-elimination noise plus small samples: “my model is good” and “my model survived this tournament” stayed different sentences.

Next focus

Carry the discipline into summer projects without chasing leaderboard ghosts.

<!-- status: backfill week_id: 2026-W04 evidence_tier: strong source_cluster: HPC cache research, Prof. Yuxi Hong group -->

Week of 2026-01-19

Week range

2026-01-19 through 2026-01-25.

Headline summary

I joined Prof. Yuxi Hong’s group and started treating Monte Carlo runs on BigRed200 like evidence problems, not vibes. The week was mostly structure: Slurm, queues, and the embarrassment of plots that only look good at small scale.

Technical work

• Cluster jobs for finance-heavy Monte Carlo work; iterated on directory layouts so figures trace back to run IDs (see portfolio/hpc-cache-research.md and published blog hpc-cache-evidence-pipelines).
• Learned again that “cache looks fine” is not a result.

Projects progressed

• HPC cache research: Evidence bundle rules; pairing reuse-distance with locality only when run IDs match.

Research / academic

• Spring 2026 includes High Performance Computing (E517) alongside lab work (per canonical profile).

Career / leadership / recruiting

• Not the focus this week.

Reading / papers / books

• Lab-internal reading on measurement; nothing public to cite without advisor alignment.

Personal development

• Patience with slow feedback loops on the cluster.

Wins

• A full-budget run at the scale I discuss publicly (548 pricings) became the anchor number I use when explaining “full width” problems.

Friction

Node heterogeneity and queue timing made early graphs misleading. Fixed with discipline, not a new color palette.

Next focus

Tighten narrative and outputs toward publication-quality artifacts, not just finished jobs.

<!-- status: backfill week_id: 2025-W37 evidence_tier: moderate source_cluster: IU Fall 2025 start, transfer credits -->

Week of 2025-09-07

Week range

2025-09-07 through 2025-09-13.

Headline summary

Fall term at IU Bloomington in Intelligent Systems Engineering was underway. I was stacking math and systems coursework on top of SCAD transfer credits, not pretending the path was linear.

Technical work

• Engineering Computing Architecture and Software Systems Engineering (summer carryover completed per transcript snapshot in canonical profile).
• Fall 2025 load included Computer Systems Engineering, Intelligent Systems I, Physics 1, Engineering Ethics, Career Development for ISE (grades on file in CANONICAL_PROFILE; not for public flex).

Projects progressed

• Personal tooling and repo hygiene on GitHub (~/Documents/GitHub/ allowlist) without claiming shipped products from class assignments alone.

Research / academic

• No BigRed200 research yet; lab alignment came later (Jan 2026 per portfolio).

Career / leadership / recruiting

• Career Development for ISE course active; resume iterations in allowlisted resume folders.

Reading / papers / books

• Course texts and syllabi only; no public book list this week.

Personal development

• Adjusting to Bloomington cadence after SCAD transfer.

Wins

• Solid grades in several Fall 2025 courses (see canonical profile table; aggregate only in public copy).

Friction

Physics 1 was the rough edge (later grade C+ in profile). I did not pretend every core was equally strong.

Next focus

Tighten fundamentals before taking on HPC-sized measurement stories.

<!-- status: draft week_id: YYYY-Www evidence_tier: summary | moderate | strong -->

Week of YYYY-MM-DD (Monday)

Week range

YYYY-MM-DD through YYYY-MM-DD (America/Indiana/Indianapolis).

Headline summary

<!-- One paragraph. What moved. -->

Technical work

Projects progressed

Research / academic

Career / leadership / recruiting

Reading / papers / books

Personal development

Wins

Friction

<!-- What was harder than expected. -->

Next focus

Timeline module

This folder contains the weeknotes system for Phase-3.

Structure

• drafts/weeknote_2026_W14.md - current weeknote draft with selected entries
• publish_ready/weeknote_2026_W14.json - structured payload for Wix or manual import
• yearly files in timeline/2025/ and timeline/2026/ - backfill and current week records

Publishing

• Preferred: map into a CMS collection (WeeklyUpdates) when collection setup is available.
• Fallback: publish weeknote as a Blog "Live Updates" post and cross-link from timeline page.

Timeline (living record)

Last updated: 2026-03-31

This folder holds weekly and backfilled timeline entries: short, sectional updates that show motion (builds, learning, friction, next focus). They are not full essays.

How to read

• YYYY/week_*.md — one file per week (Monday date in filename).
• Backfill — older weeks populated from the past-year miner (content_ops/backfill_manifest_*.json).
• Current week — latest draft ends the list until published.

Index (chronological)

Week file	Topic (short)
2025/week_2025_09_07.md	Fall 2025: IU engineering stack, transfer term
2026/week_2026_01_19.md	HPC lab: BigRed200, evidence bundles
2026/week_2026_02_09.md	NCAA Final Four Analytics Challenge
2026/week_2026_03_09.md	Claude hackathon telehealth prototype
2026/week_2026_03_16.md	OpsPilot / workflow agent, contracts over vibes
2026/week_2026_03_23.md	Website_Per content pipeline + publish readiness
2026/week_2026_03_31.md	Current week (Phase 3 system layer)

Publish path

• Site page draft: pages/timeline.md
• Wix blocks: publish_ready/timeline_blocks.md
• API: usually Editor-only; record rows in content_ops/publish_status_registry.json