Month 11-Week 1: The capstone long-form post¶

Week summary¶

Goal: Write the post-the long-form (3500–5000 word) technical writeup of your capstone. The artifact that hiring managers read top-to-bottom.
Time: ~10 h over 3 sessions.
Output: Drafted, edited, and reviewed long-form post-ready to publish next week.

Why this week matters¶

Engineers undervalue distribution. The capstone post is what makes the year legible to people who didn't watch you build it. Done well, it pays career dividends for 2–3 years. The bar for this post is higher than any prior-it should be referenceable in your CV.

Prerequisites¶

M10-W04 complete with v0.1.0 shipped.

Recommended cadence¶

Session A-Tue/Wed evening (~3 h): outline + first 1500 words
Session B-Sat morning (~4 h): finish draft + edit
Session C-Sun afternoon (~3 h): review + polish

Session A-Outline + first sections¶

Goal: Detailed outline. Draft motivation + problem + approach.

Part 1-Outline (60 min)¶

Aim for 4000 words. Structure:

1. Hook (250 w)
   The specific problem; the specific quantified result; tease the depth.
2. Problem framing (500 w)
   Why this matters in 2026 LLM systems.
   What real teams currently do.
   What's missing.
3. Existing tools-honest survey (500 w)
   For each major incumbent: what it does well, what it lacks.
4. The approach (700 w)
   Architecture sketch.
   Key design decisions and why.
   Code snippets (well-chosen, not too many).
5. Results (700 w)
   Eval setup (dataset, metrics).
   Headline numbers with bootstrap CIs.
   Comparison vs incumbents (table).
   Failure modes.
6. What I learned (500 w)
   Technical insights.
   Engineering insights.
   Specialty-meta-insights.
7. Limitations + what's next (300 w)
   Honest gaps.
   Roadmap.
8. Closing (250 w)
   Bridge from the year's narrative.
   Call to action (try it / contribute).
   Links: repo, results, prior posts in the series.

Part 2-Hook + Problem + Approach (90 min)¶

Write sections 1, 2, 4 (skipping 3 for now). ~1500 words.

Part 3-Set aside (30 min)¶

Save. Don't re-read tonight. Sleep on it.

Output of Session A¶

Detailed outline.
~1500 words of draft.

Session B-Finish draft + first edit¶

Goal: Complete the draft. First edit pass.

Part 1-Existing tools survey (45 min)¶

Section 3. Be specific and fair. Avoid uncharitable readings.

Part 2-Results section (90 min)¶

Section 5 needs: - Charts (loss curves, eval scores, comparison bars). - Tables (the numbers). - Bootstrap CIs. - Honest failure-mode breakdown.

Part 3-What I learned + limitations + close (60 min)¶

Sections 6, 7, 8.

Part 4-First edit pass (45 min)¶

Read aloud. Tighten. Cut filler.

Output of Session B¶

Complete draft (~4000 words).
First edit pass done.

Session C-External review + polish¶

Goal: One outside reader. Apply feedback. Final polish.

Part 1-Find a reader (30 min)¶

Ask 2 people: - A peer in your specialty. - Someone outside the specialty (tests if the post is accessible).

Send them the draft + 4 questions: 1. Does the hook make you want to read on? 2. Did anything confuse you? 3. Is anything overly hyped or under-hyped? 4. Would you forward this to a friend?

Part 2-Apply feedback (75 min)¶

Address every substantive comment. Often: - Hook needs sharpening. - Approach section needs more diagrams. - Results need more context. - Closing falls flat.

Part 3-Final polish (75 min)¶

Read aloud.
Verify every number / link.
Format code snippets cleanly.
Image alt-text.
Title and subtitle (the title bar of a tweet you'll write next week).

Output of Session C¶

Polished, externally reviewed long-form post.
Ready to publish next week.

End-of-week artifact¶

4000-word capstone post drafted, edited, reviewed
Charts and code snippets in place
At least one outside reader reviewed

End-of-week self-assessment¶

My post is something I'd link in a job application.
My post would survive a critical read by someone in my specialty.
I'm ready to broadcast next week.

Common failure modes for this week¶

Polishing forever. Done is better than perfect.
Skipping the outside reader. Yours-only is too biased.
Defensive wording. Confidence + honest limitations beats hedging.

What's next (preview of M11-W02)¶

Publish broadly. Give a talk (internal first; external as stretch). Engage with feedback.