SIH26038 — explainable AI screening for diabetic retinopathy, built for rural India. This maps the six-slide script onto what actually goes on screen, with the real Day-1 prototype numbers already dropped in.
6 slidesPS ID SIH26038
SoftwarePS category
Day-1prototype results, not projections
NetraCore — SIH26038Design guide — 02 / 07
What the template actually requires
The script already fits the format. Here's the exact rulebook it has to match, pulled from the official template's own instruction slide — plus the title-slide fields the brief left out.
Six rules from the template
Six slides total, title included. Title, idea, technical approach, feasibility, impact, references — you're exactly at six. Don't add a seventh.
Points and visuals only. No paragraph blocks — every line should be a bullet, a diagram, or a picture.
Precise, easy to follow. If a judge can't get the point on one read, cut it.
The idea has to read as novel. That's the whole job of the comparison table on Slide 2.
Keep the mandated content pointers. The template locks the bullet categories per slide (listed on each spec page ahead) — restyle freely in Canva, just don't drop what it asks for.
Export as PDF, not PPTX. The portal only accepts PDF. In Canva: File → Download → PDF Standard.
Title slide — required fields
Field
What goes there
Problem Statement ID
SIH26038 — confirmed
Problem Statement Title
Missing — pull the exact wording from your PS listing on the portal
Theme
"Clean & Green Technology" — confirm against the portal listing
PS Category
Software
Team ID
Missing — add before export
Team Name
NetraCore — confirmed
Easy to mix up
Theme and PS Category are two separate fields on the official slide — the working draft only accounted for one of them (it called "Clean & Green Technology" the PS category, but that's the Theme; PS Category is simply Software or Hardware). Fill in both, separately.
NetraCore — SIH26038 Canva build guideCross-check against the Q&A fact sheet before export
NetraCore — SIH26038Design guide — 03 / 07
Brand kit — set this up first
Build this once as a Canva Brand Kit and every slide inherits it. Ten minutes now saves an hour of six slides drifting into six different decks.
Palette
Forest
#1A4D2E
Primary — headers, solid fills
Forest Deep
#0F3320
Dark backgrounds, dividers
Sage Tint
#EAF1EC
Light card backgrounds
Amber
#C9A227
Accent only — badges, callouts
Ink
#16241C
Body text
Paper
#FFFFFF
Base background
Typography — both free in Canva's font library
Poppins — headings, labels, data numbers (Semibold / Bold)
Explainable screening, built for rural India
Rounded geometric sans — carries the "healthcare for people," not enterprise-software, feel
Lora — body copy, bullet text (Regular / Semibold)
A retinal photo, graded 0–4 on the ICDR scale, becomes a referral decision a field worker can act on.
Warm serif, ≥20pt on every slide — judges read from a distance
Canva workflow tip
Build Slide 1's header/footer corner marks once — team name, slide number — then Duplicate page for every new slide and swap only the body content. That's what keeps the corners pixel-identical across all six; rebuilding them by hand each time is where decks start to look inconsistent. And skip stock photography entirely — if Slide 1 wants a human image, it's the farmer from the spoken hook, as a respectful illustration, not generic medical stock art.
NetraCore — SIH26038 Canva build guideSet the Brand Kit before opening Slide 1
NetraCore — SIH26038Design guide — 04 / 07
1
Slide 1 of 6 — Title Page (official header) · Tanzil, ~1 min, right after the spoken hook
Uncluttered — the emotional weight already landed
Required on this slide
Problem Statement ID, Title, Theme, PS Category, Team ID, Team Name — field table on page 2
Team name as the visual anchor
Tagline: "Explainable AI screening for diabetic retinopathy, built for rural India."
This slide plays right after Tanzil's spoken hook about the farmer in his 50s — the story has already landed by voice. Let the six required fields sit small and factual; let the team name and tagline do the only visual work.
PS ID · PS Title · Theme · PS Category · Team ID
NetraCore
Explainable AI screening for diabetic retinopathy
Canva build
One text box, six short lines, small Lora, top strip. Team name in big Poppins Bold, centered, is the only element that should draw the eye.
2
Slide 2 of 6 — Idea Title (official header) · Nidhi (problem → solution), then Mayank (uniqueness)
Two halves, one slide: solution, then difference
Required · Half A (Nidhi)
Proposed solution, detailed explanation, how it addresses the problem, innovation & uniqueness
Retinal photo (low-cost cameras)
→
0–4 ICDR grading
→
Referral decision + explanation
No numbers needed here — Slide 5 carries the burden stats.
Half B — uniqueness (Mayank)
Diagnostic grading
Explainability
Capacity modeling
IDx-DR / EyeArt
✓
✗
✗
Remidio / Forus
✓
✗
✗
NetraCore (ours)
✓
✓
✓
Canva build
Table element, 4×4 — check = Forest tick icon, ✗ stays muted ink-grey (a gap, not an error). Match the flow boxes' corner radius to the table cells so the two halves rhyme.
NetraCore — SIH26038 Canva build guideSlides 1–2 of 6
NetraCore — SIH26038Design guide — 05 / 07
3
Slide 3 of 6 — Technical Approach (official header) · Tanzil, ~4 min — slide is the backdrop, the live demo is the event
The pipeline, then the live app
Required: technologies used · methodology / implementation
Image Input
Quality Check 84.3% pass rate
Contrast Enhancement (CLAHE)
ResNet-18 Classifier
Grad-CAM Explainability
Referral Decision
Don't print these here
Skip 84.31% sensitivity / 94.96% specificity — Tanzil says them out loud right after the demo. They belong on Slide 5.
Backup screenshots (contingency)
No DR — 99.97% confidence, "NO REFERABLE DR"
Severe DR — 77.64% confidence, "REFER TO SPECIALIST," Grad-CAM heatmap on the hemorrhage region
Canva build
Pin the pipeline as a static frame behind the live app — don't animate it. Annotate confidence % and grade directly onto both backups so the fallback path looks identical in quality to a live run.
4
Slide 4 of 6 — Feasibility and Viability (official header) · Aryan, ~1.75 min
Centerpiece: dual-axis line chart from run_capacity_sweep.m — x: tele-ophthalmologists (0–5) · left y: backlog after 1 year (images) · right y: utilization. Drop in the actual figure, not a redraw.
Connectivity
Offline capture + batch upload
Power
Intermittent — designed around it
Operators
Existing field health workers
Existing-infrastructure strip
eSanjeevani — 430M+ consultations
Remidio smartphone fundus camera
Forus 3nethra
Logos or labeled icons, not paragraphs — reinforces "no new hardware to buy."
Canva build
Canva's chart element can't do a true second y-axis — don't fake it there. Export the real MATLAB figure as a PNG in a thin bordered frame. Pull all three icons from one Canva set — don't mix glossy 3D with flat line style.
NetraCore — SIH26038 Canva build guideSlides 3–4 of 6
NetraCore — SIH26038Design guide — 06 / 07
5
Slide 5 of 6 — Impact and Benefits (official header) · Nidhi, ~1.5 min — also home for the honest model numbers
Human-scale burden, then the honest numbers
Required: impact on target audience · benefits
101M
people with diabetes in India — highest in the world
~17%
DR prevalence among them (Govt. National DR Survey)
15/M
ophthalmologists nationally, vs. WHO's 1-per-50,000 target
Referable-DR confusion matrix (n=462)
Pred. +
Pred. −
Actual +
TP 172
FN 32
Actual −
FP 13
TN 245
Model performance — target vs. achieved
90%
84.31%
Sensitivity
85%
94.96%
Specificity
5-class accuracy: 75.54% — reported for completeness; sensitivity/specificity are the clinically meaningful numbers.
Canva build
Hand-build the bars with rectangles sized to the percentages — outline for Target, solid amber for Achieved. Canva's native chart tool can't mix outline and filled series, and that contrast is what makes the honesty legible.
6
Slide 6 of 6 — Research and References (official header) · Manya, ~30 sec
Clean, scannable, done in 30 seconds
Required: details / links of reference and research work
Reference architecture — MathWorks' official diabetic retinopathy classification example (ResNet-18, transfer learning)
Cited prior work — IDx-DR / LumineticsCore, EyeArt, Remidio Medios AI, Forus 3nethra; ICDR Severity Scale; Selvaraju et al., Grad-CAM
Canva build
Small type, three clean groups, generous line spacing — no chart needed. Manya only has 30 seconds, so this slide shouldn't hand her anything to explain beyond reading it aloud.
NetraCore — SIH26038 Canva build guideSlides 5–6 of 6
NetraCore — SIH26038Design guide — 07 / 07
One more moment worth adding
The script leans hard into "we tell judges what's real vs. proxy vs. aspirational" as a credibility strategy. It deserves a small visual moment of its own — a compact card near the close, or a persistent footer tag — not a full extra minute of speaking time.
Real & validated tonight
Trained classifier, real Grad-CAM, working quality gate, Simulink capacity model
Lightweight proxy (said out loud)
Lesion highlighting is morphological filtering, not a trained segmentation network
Aspirational
Hitting the full >90% / >85% target needs more data and training time — Slide 5 shows where you actually are
Final build checklist
Re-run run_capacity_sweep.m and export the real Simulink chart image for Slide 4
Grab clean, annotated screenshots of both demo cases (No DR @ 99.97%, Severe DR @ 77.64% with Grad-CAM) for Slide 3's backup path
Build the competitor comparison table for Slide 2
Hand-build the target-vs-achieved bar chart for Slide 5 — not Canva's native chart tool
Set up the Brand Kit — palette + fonts — before touching individual slides
Fill in Problem Statement Title, confirm Theme, and add Team ID on Slide 1
Cross-check every number on every slide against the Q&A Survival Guide fact sheet
Confirm the slide count is exactly six, then File → Download → PDF Standard
NetraCore — SIH26038 Canva build guideBuilt from the Day-1 prototype results