Siddhartha Manaktala’s grandmother lived with diabetic retinopathy for years. He and his dad both wear strong corrective prescriptions. At 17, still in Class XII in Gurugram, he surveyed more than 200 low-income and elderly neighbours — and found that only three of them had ever seen an ophthalmologist. He built EyeSpy, an AI smartphone platform that screens for seven eye diseases, then took it into community eye camps instead of waiting for people to find a clinic. Reporting: Times of India / ET Edge Insights (2026).

How AI actually helped this project succeed (facts only)

This was not “ask a chatbot for the answer.” The published pieces of the system that matter:

1. Seeing the eye from a phone. EyeSpy uses a phone camera to capture images, then runs two AI pipelines — one for the outer eye and one for the retina. No specialized clinic equipment is required (an optional lens attachment costs about Rs 1,200). On an entry-level Android phone the screen runs in under three seconds and can work offline.
2. Classifying seven diseases. The platform screens conditions including cataracts, conjunctivitis, pterygium, diabetic retinopathy, and glaucoma — seven diseases in total. The first AI model hit only 71% accuracy at an eye camp. Manaktala spent a week at L V Prasad Eye Institute in Hyderabad with doctors, researchers, and the tech team, then rebuilt the dual-pipeline system.
3. Getting screening to the people who miss clinics. Field camps across rural and suburban Haryana screened more than 1,650 people: 99% had never had an eye exam; over 25% had undiagnosed eye disease; over 35% did not know they had a vision problem. Across four camps, accuracy vs attending ophthalmologists reached 90.1%. When people could not afford prescriptions, he crowdfunded Rs 1 lakh for field kits (glasses, drops, care instructions). One camper, Asha, 12, had been sitting at the back of class thinking she was bad at school — she needed glasses.

Also on the record: silver awards at the IRIS and INSEF national science fairs, with mentorship from LVPEI and Lenskart.

So AI’s job here was specific: turn a phone photo into a disease-class decision fast enough for a village camp, then rebuild until the decision matched specialists well enough to trust in the field.

What CodeSky kids practice when we teach from cases like this

At our Markham afterschool we do not ask kids to copy a week inside an eye institute. We do ask them to practice the same computational thinking shape EyeSpy used:

• Decomposition — split “screen eyes in a village” into see → decide → finish the mission.
• Abstraction — treat the camera as a sensor that hands the machine an image, not a magic diagnosis; treat “which of seven labels?” as a rule with a measurable accuracy score.
• Algorithms — write a loop the field team can finish: capture → classify → kit or referral, including the offline path when the network is gone.
• Pattern recognition — rebuild the classifier when 71% is not good enough (the published jump to 90.1% is the outcome of that work, not a classroom guarantee).

In our classroom language that maps to see with a sensor → judge with a rule → run a mission (phone camera → AI classify seven diseases → village camp). A kid who can name those three steps can use AI as a tool inside a system they still own — not as a substitute for thinking.

Trial invite: walk-in coding afterschool in Markham. 9833 Markham Rd Unit 2, Markham, ON L6E 0E5. Email info@codesky.ca.

Follow CodeSky

• Instagram: @codeskyca
• 小红书 (Xiaohongshu / RedNote): CodeSky profile
• WeChat 公众号: search codeskyca

Sources

• Times of India / ET Edge Insights (2026), coverage of Siddhartha Manaktala and EyeSpy (dual AI pipelines, seven diseases, 71% → 90.1%, <3s offline Android, Haryana camps, 1,650+ screened, Asha, Rs 1 lakh kits, IRIS/INSEF silver, LVPEI + Lenskart).