Table of Contents
MedReport AI
MedReport AI is a product I built for a problem I face personally. As a diabetic with ankylosing spondylitis, I receive blood reports every quarter. The first time I saw my HbA1c value on a piece of paper with no explanation, I spent twenty minutes on Google, and left more confused than when I started. This case study covers how I designed MedReport AI, a tool that translates clinical lab output into plain language that patients can actually act on.
The Problem
India runs over 2 billion diagnostic tests every year. When patients receive their results, they are handed a document dense with clinical abbreviations, reference ranges, and jargon that most cannot interpret. Without a doctor callback, which averages 3.2 days, they either search Google or wait in anxiety.
Most people in India do not have a GP they can call the same afternoon. They paste reports into WhatsApp groups, search US-centric medical sites with incorrect reference ranges, or simply fold the paper and hope everything is fine. None of these are acceptable for a country with 101 million diabetics who need to understand their numbers regularly. The problem is not the report itself. It is the translation layer that does not exist between clinical language and patient comprehension.
The Solution
MedReport AI accepts uploaded PDF reports from India’s top five diagnostic labs like SRL, Dr. Lal PathLabs, Thyrocare, Metropolis, and Apollo. It extracts key findings using a Gemini prompt chain and returns a structured plain-language summary in under 30 seconds. Abnormal findings are flagged with severity context. Normal findings are confirmed. Every output ends with a recommended next action
The design principle throughout was confident clarity, never alarming, never vague. Every output was required to tell the user exactly what to do next, even when all results were normal.
You can click on respected Tabs Below to see some innovative info and features!
Hb: 9.2 g/dL (L) | MCV: 78 fL (L) | RDW: 16.2% (H). Suggest iron studies and peripheral smear.
Your red blood cell level is lower than normal, which can cause tiredness and breathlessness. Your doctor will likely want to check your iron levels with a follow-up test. Book an appointment this week.
Process Intro
The build had four phases. Each one surfaced a decision that shaped the product in ways I did not anticipate at the start.
Results Intro
Prototype testing across 12 users showed consistent improvements in comprehension speed and confidence.
Validate tone before building extraction logic
Spent two weeks on PDF parsing before testing if patients preferred plain language. A 30-minute copy test first would have confirmed the assumption. Validate the output then build the mechanism.
Prominent disclaimers increase trust, not decrease it
Early versions buried the disclaimer at the bottom. Users treated AI output as diagnosis. Moving it to a prominent card at the top increased trust scores from 6.2 to 8.4 out of 10 in prototype testing.
The all-clear state is a full product moment
When all results are normal, everything looks fine is not enough. The best version ended: Your results are within normal ranges, your next routine check-up is sufficient. That closure matters as much as flagging abnormals.
India has 101 million adults living with diabetes, the second highest count globally. IDF Diabetes Atlas, 2021.
Explore AI photography techniques, brand imagery case studies, and copy-ready Gemini prompts.