JETRO 3-min Pitch Video — Kashi Production Briefing

⚠ Two-program timeline

Alchemist Enterprise B2B Course — 2026-05-31 23:59 JST (tomorrow). Submit: English entry form + the 10-page deck. No video required. Send deck-en-10slide.html PDF link in the deck field. ⚠ Add your LinkedIn URL to S9 first.
J-StarX SV Extended Program — 2026-06-22 17:00 JST. Submit: English entry form + the same 10-page deck + the 3-min English video as an Unlisted YouTube/Vimeo link.
Set the video to "Unlisted", NOT "Private" — Private means the committee can't open it.
Test the link in a private/incognito window before pasting it into the form.

Pre-submit checklist (run after recording)

Final cut runs ≤ 3:00 (v3 script targets 2:44 at 130 wpm — you have ~15s of headroom for natural pauses)
Audio is clean, no clipping; face is lit from the front; eyes look at the lens
Lower-third text cards appear cleanly at the [CARD: …] marks in the script (see cue map below) — readable for ~3 seconds each, not blocking your face
Every spoken number matches canon (908 tests / 50 meetings / 5,000 turns / 5-person minimum) — no invented customers or pilots
Opening lower-third name card is correct (Justine Acaylar · Founder, Kashi)
Uploaded Unlisted, link opens in incognito, pasted into the right J-StarX field
10-page deck PDF uploaded separately and linked in both program forms

The Script v3 · Option B: talking head + lower-third text cards

357 words · target pace 130 wpm = ~2:44 with ~15s pause headroom under 3:00. Camera stays on you the whole time. The CARD: … mark tells the editor when to drop in a text overlay (lower-third or full-bleed). 6 cards total over 3 minutes — visual punctuation, not slides. Teleprompter-formatted version: script-teleprompter.html.

1 · Hook / Cover + Problem0:00 – 0:30 · ~64 words

Before anyone says a word, trouble in a team shows up. I spent six years on a small company's board in Japan, watching good teams quietly fall apart. Every survey came back saying things were fine. By the time anyone spoke up, the best people had quit. The signs were visible long before — in how the meetings ran. But no tool could see them.

CARD (lower-third, 0:00–0:05): "Justine Acaylar · Founder, Kashi"

2 · What existing tools miss0:30 – 0:50 · ~37 words

I'm building Kashi. In Japan, preventing workplace harassment is now a legal obligation — even for small firms. But it stays invisible until people leave. Surveys, hotlines, even meeting AI — every existing tool depends on someone saying something.

CARD (full-bleed, ~0:30, 2 sec): "Kashi" wordmark

3 · The insight + solution0:50 – 1:20 · ~47 words

Kashi doesn't judge what was said. It reads the shape of how a team interacts — who talks, who gets interrupted, whose questions go unanswered. In one meeting it looks like noise. Across many meetings, it's a pattern. That's what we surface — over 30, 90, and 180-day windows.

CARD (lower-third, ~0:55, 3 sec): "Reading the SHAPE, not the content."

4 · Three views + boundaries1:20 – 1:45 · ~46 words

What you see depends on your role. Executives see which teams need attention. Managers get a private mirror of their own team. Members see only themselves. Meeting content, scores, emotion — none of that is shown. The limits are built into the data — they can't be turned off.

CARD (lower-third, ~1:38, 3 sec): "Limits built into the data — not the screen."

5 · Why now + Moat1:45 – 2:15 · ~60 words

And the timing is sharp. Japan's harassment law just expanded to small firms. The EU AI Act, Article 5, now restricts workplace emotion-recognition. Tools that judge from content face real exposure. Kashi sits on the other side of that line — rule-based, deterministic, no emotion or sentiment. Content is never stored for evaluation. Hard to copy without rebuilding the whole product.

CARD (lower-third, ~1:55, 3 sec): "EU AI Act · Article 5"

6 · Traction + Team2:15 – 2:40 · ~55 words

I'll be straight. We're pre-pilot, with no paying customers yet. But the analyzer works — 908 tests, including a determinism lock. On simulated meetings, the patterns came through cleanly, with zero false positives. Stress-tested at 50 meetings, 5,000 turns. I'm Justine Acaylar — at Keio, building solo, Filipino-Japanese. Exactly the multicultural dialogue this is built for.

CARD (lower-third, ~2:18, 3 sec): "908 tests · determinism lock"

7 · Ask + Closer2:40 – 3:00 · ~48 words

Kashi isn't surveillance software. It's trust infrastructure — built to travel to US and EU buyers. From an acceleration partner, I'm asking for three things: US market-fit validation, large-enterprise PoC introductions, and AI-governance mentoring. We never judge what your teams say. We help them hear how they say it.

CARD (full-bleed text, ~2:55, 5 sec): "We never judge what your teams say. / We help them hear how they say it."

Production Runbook (talking-head + teleprompter)

Total hands-on time ~2–4h. Record in the 7 sections so any flub is a 20–30s redo, not a full restart.

1 · Set up the teleprompter (free)

Easiest, free, no install: open cueprompter.com or a free web teleprompter in a browser, paste the plain-text block from script-teleprompter.html (bottom of that page), set speed slow.
Position it directly under or beside the camera lens so your eyes stay on-lens. If the prompter is off to the side, it reads as “looking away.”
iPhone option: apps like BIGVU or Teleprompter (App Store) let the phone be both camera + prompter; the script scrolls over the lens.
Set scroll speed to ~150 wpm and do one practice pass to tune it.

2 · Camera, audio, lighting

Camera: iPhone (rear camera, 1080p, landscape) on a tripod beats the MacBook webcam. Frame chest-up, eyes in the upper third.
Audio (most important): quiet, soft-furnished room. External USB or lav mic if you have one; otherwise keep the phone/mic within ~50cm. Record a 5-second test and play it back before the real take.
Lighting: face a window or a lamp — light source in front of you, never behind. No backlit window.
Background: clean wall, or a tidy bookshelf. Nothing distracting.

3 · Record in 7 sections

Record each script section as its own clip. Pause, breathe, restart a clip if you stumble.
Leave ~1s of silence at the start and end of each clip (clean edit handles).
Do 2 takes of the Hook (section 1) and the Ask (section 7) — those bookends matter most.

4 · Edit (CapCut — free, fastest)

Import all 7 clips into CapCut (or iMovie). Stitch in order.
At each CARD: … cue, add a text overlay matching what the script specifies:
- Lower-third cards (Justine Acaylar / Reading the SHAPE / Limits in the data / EU AI Act / 908 tests): bottom-left, ~3 seconds, 32-40pt sans bold, white on a subtle dark band. Don't cover your face.
- Full-bleed cards (Kashi wordmark at ~0:30; final tagline at ~2:55): centered, large font, ~2-5 seconds. The final tagline card is the closer — let it land.
Optional: burn in English captions (CapCut auto-captions, then proofread — fix mis-heard terms like "Kashi", "MHLW", "APPI", "Acaylar").
Trim dead air between clip seams. Confirm total ≤ 3:00.

5 · Export & upload

Export 1080p, H.264 MP4, 30fps.
Upload to YouTube → Visibility = Unlisted (NOT Private). Title e.g. “Kashi — JETRO Pitch — Justine Acaylar.”
Copy the link, open it in an incognito window to confirm it plays, then paste into the JETRO form.

Delivery Coaching

Pace: ~130 wpm — slower than v2, deliberately. Lands at ~2:44 with ~15s of pause headroom. Use the time. Pause a full beat at every "—" and at each section break.
Your English is an asset. Don't flatten your accent or over-formalize. A confident Filipino-Japanese founder is exactly the credible voice for "multicultural dialogue." Speak like you'd explain it to a smart friend who doesn't already know your industry.
Land the bookends. The opening line ("Before anyone says a word, trouble in a team shows up.") and the closing line ("We never judge what your teams say. We help them hear how they say it.") are what the committee remembers. Slow down on both — pause after "word," and after "say."
Eyes on the lens, not on the scrolling text. The prompter should be ~at lens height so the difference is invisible.
One phrase to enunciate clearly: "EU AI Act, Article 5." It's the regulatory beat.

Why this script (so you can defend any line)

Choice	Reason
Option B (talking head + text cards) for the video	Per Alchemist insider guides + J-StarX applicant videos: the application screen is about the founder, not slide-heavy product explainers. Pure delivery + occasional text accents shows you can sell — the core B2B founder test. (Full slide cutaways are more of a Demo Day style.)
10-page deck (v4) as the single source for both programs	Enterprise B2B Course asks for "approximately 10 pages"; J-StarX has no page minimum. One deck covers both — no version drift, no extra maintenance.
Cover line "Before anyone says a word" opens the spoken hook	Echoes the cover slide — burns the message in immediately and pays off again on the closer line.
Founder-origin story still anchors Section 1	The 3-judge panel ranked this highest: the 6-years-on-a-board story earns your right to the pitch before any product claim.
"Built into the data itself, not into the screen"	Replaces the v1's "database layer" jargon. Same meaning, plain language.
"From an acceleration partner" (not naming a specific program)	Program-neutral phrasing — script + deck reusable across J-StarX, Enterprise B2B, and any other applications.
No hard TAM/ARR number spoken	Conservative per investor canon (TAM/SAM/SOM are hedged pre-pilot). The deck carries the figures with "model" qualifier; the video stays defensible.
"pre-pilot, no paying customer logos"	Honest traction is canon. Never claims customers/pilots that don't exist (2026-05-28 fabrication lesson).
"designed to align with"	Avoids the banned "fully compliant" / "legally safe."
"We never judge what your teams say"	Closer is accurate (Kashi never judges on content) and avoids the banned "we do not transcribe" sleight-of-hand.

Full line→source map + the forbidden-phrase grep log: 00-script-canon-audit.md.

Slide × Script Timing Map (JA 16-slide deck → 3-min EN video)

Deck used = Kashi_PitchDeck_v3_Final_2026-05-27.pdf (16 JA slides). The 3-min EN script uses 10 of them as cutaways; 6 are deliberately skipped (reasons at the bottom). Edit in CapCut: drop each PDF page as an image overlay at the listed timestamp; the rest of the time your face is on camera.

Time	Slide	What you say at that moment (key phrase)
0:00 – 0:10	01 Cover	"For six years, I sat on the board of a multinational small company in Japan."
0:10 – 0:30	02 Problem	"Meeting by meeting, I watched good teams quietly fall apart. Every survey came back without a flag. By the time anyone spoke up, the best people had already quit."
0:30 – 0:45	04 気づき	"I'm building Kashi. In Japan, preventing workplace harassment is now a legal obligation — even for small firms, under the MHLW law. But it stays invisible until people leave… every existing tool depends on someone speaking up, or on what they say."
0:45 – 1:00	05 Solution	"Kashi doesn't judge what was said. It reads the shape of how a recurring team interacts across many meetings: turn timing, who gets interrupted, whose contributions get no response."
1:00 – 1:13	14 Why Now	"Here's the shift. Japan's mandate just expanded to small firms, and the EU AI Act — Article 5 — restricts workplace emotion-recognition. So tools that read content or infer sentiment now face real exposure."
1:13 – 1:25	13 差別化	"Kashi was designed to sit on the other side of that line: deterministic, rule-based detectors, no language model in the judgment path, surfacing patterns over 30-, 90-, and 180-day windows."
1:25 – 1:45	06 Product	"The output is a private mirror to the manager themselves, plus aggregate patterns — five people or more — for executives. Content-reading tools can't reach that posture without redesigning their products."
1:45 – 2:05	07 見えるもの / 見せないもの	"Four pillars, all built for regulatory defensibility. One — structural, not emotional… Two — self-reflection, not surveillance… Three — bounded visibility is architecture, not policy: five-person minimum, enforced at the database layer. Four — barred by design from any performance, discipline, or termination use."
2:05 – 2:25	08 現在地（N=1）	"I'll be straight: we're pre-pilot, with no paying customer logos to share. But the analyzer works — bit-for-bit deterministic, 908 tests… simulated meetings… zero false positives in the no-signal cases. Demo-tenant data runs live in production."
2:25 – 2:55	16 Team + Ask	"I'm Justine Acaylar — at Keio, building product, code, and sales solo. Filipino-Japanese: exactly the dialogue this is built for. Kashi isn't surveillance software. It's trust infrastructure… From JETRO and Alchemist, I'm asking for North-American market-fit validation and large-enterprise PoC introductions."
2:55 – 3:00	01 Cover (bookend back to wordmark)	"We never judge what your teams say. We help them hear how they say it."

Skipped — and why

Skipped slide	Why not in the 3-min video
03 既存アプローチが見逃すもの	Redundant with 02; the script already lists existing tools verbally ("surveys, hotlines, content classifiers"). Keep it in the deck for the JETRO panel to read in the PDF.
09 6 ヶ月で検証すること	This is the 1stRound-specific 6-month validation plan. JETRO ask is different (NA validation + enterprise PoC intros). Out of scope for the video.
10 初期顧客	The 3-min script doesn't explicitly state "50–500-person SMB" target. If you want this on screen, swap it in at 0:30–0:45 in place of 04 (your call).
11 市場規模	Deliberately not flashed on screen. The video stays conservative (canon hedge — TAM/SAM/SOM are "figures undisclosed pre-pilot"). The deck PDF carries the numbers; the spoken video doesn't show them.
12 ビジネスモデル	The video doesn't discuss pricing. Stays in the deck.
15 1stRound が解く 6 ヶ月の壁	1stRound-specific — wrong audience for a JETRO submission. Replace with the JETRO ask language (already in the script).

Edit shortcut

In CapCut: import the PDF (or 16 PNG screenshots — Preview ⌘P → "Save as PDF" gives you each page). For each row above, drop the matching page as an image layer at the timestamp, sized to fill maybe 70% of the frame on the right while your face stays on the left third. Or: cover full-frame, then cut back to your face for the next sentence. Keep the cover-back transitions short (≤0.3s).

JETRO Application Form — Fill-in Answers

Spec confirmed by the form itself: pitch video = within 3 minutes / English / accessible link. The 2:53 script fits. Below: drafted answers (copy-paste) + the fields only you can fill.

Product/Service Information — 200+ words English (必須)

245 words · 0 forbidden-phrase hits · Product + Issues + Target as the form asks. Copy verbatim:

Kashi is a B2B SaaS that helps organizations see dialogue breakdown in their teams before anyone has to speak up — by reading the structure of how a team interacts across many meetings, never the content of what is said.

The problem. In Japan, preventing workplace harassment is now a legal obligation, expanded to small and mid-sized firms under the MHLW パワハラ防止法. But organizational breakdown stays invisible until people quit, because every existing tool — engagement surveys, hotlines, content classifiers, organizational network analysis — depends on someone speaking up, or on what they say. The structural warning signs appear months earlier, in how meetings actually run, and no existing tool reads them.

The product. Kashi converts meeting transcripts into structural features — turn timing, interruption direction, floor-share asymmetry, response latency — and surfaces repeated patterns across 30-, 90-, and 180-day windows using deterministic, rule-based detectors, with no language model in the judgment path. The output is a private mirror to the manager themselves, plus aggregate patterns (five people or more) for executives. Raw content is not retained past a short TTL. The design is built to align with the EU AI Act (Article 5, which restricts workplace emotion-recognition) and is APPI-aware on data retention.

Target. Our initial market is 50-to-500-person, meeting-dependent firms in Japan that already use Zoom, Teams, or Meet with recording enabled, and that carry MHLW harassment-prevention obligations. We are pre-pilot today, with a working analyzer that is bit-for-bit deterministic on transcript input.

Objective of Participating (必須, multi-select)

Check these 3 (they map exactly to the video's ask):

✅ 世界へのスケールアップに向けたマインドセット / Mindset for scaling up globally
✅ 海外でのPoC検証、顧客フィードバックの獲得 / Overseas PoC validation, obtaining customer feedback
✅ メンタリングや投資家・潜在顧客とのコネクション形成 / Mentoring and forming connections with investors and potential clients
○ (optional) パートナー探し / Partner Search — fine to add
✗ 販路開拓 / Sales channel development — skip; you're pre-pilot, "PoC validation" is the honest version of this

Program Participant Information (1)

Field	Answer
氏名（日本語）/ Name (JA)	アカイラル・ジャスティン (adjust to your registered katakana)
氏名（英語）必須 — include prefix	Mr. Justine Acaylar
役職名（日本語）必須	創業者（ファウンダー）
役職名（英語）必須	Founder
Email 必須 / re-enter	⚠ your email
電話番号必須	⚠ your phone
LinkedIn	⚠ your LinkedIn URL
Facebook (optional)	⚠ optional

Other fields

ピッチ動画URL / Pitch video link — paste your Unlisted YouTube link once recorded.
ピッチデック（英語）URL / English pitch deck link — use the full 16-slide English deck (parallels your submitted JA deck 1:1, with Slide 15 rewritten for the JETRO/Alchemist ask). Export: ⌘P → Save as PDF → Layout: Landscape (each slide = 1 page) → upload to Drive/Dropbox → paste the shareable link. (The shorter 9-slide deck is for video cutaways only.)
海外拠点の所在地 / Address of Overseas Office (if any) — leave blank. You have no overseas entity yet. Don't invent one. (The Philippines is a future plan, not an existing office — say it in the video, not in this field.)
受賞歴 / Awards — leave blank unless you have a real award. Don't fabricate.
ホームページ（英語）URL — you've used the same kashi.ai-labo.workers.dev for JA and EN. Fine if the site is bilingual or EN-readable; if it's JA-only, that's acceptable here (the field just needs a working link).