ReportSimple

Track: AI + Healthcare

Inspiration

Most people get lab reports full of numbers, units and abbreviations with no explanation. They wait days for a doctor, search confusing information online, or panic over normal values. This is harder for people who are not comfortable with medical English. We wanted to build something that makes a report understandable in minutes, in English or Bangla.

What it does

ReportSimple lets you upload a lab report (PDF, photo or text) and gives you:

• a plain-language summary
• every test marked high, low or normal, with a simple meaning
• an urgency level (routine, see a doctor soon, urgent)
• questions to ask your doctor

It never diagnoses or suggests treatment, and always shows a disclaimer.

How we built it

1. OCR: Tesseract and pdf2image read photos and scanned PDFs, pdfplumber reads digital PDFs.
2. LLM extract: a Featherless AI model (Qwen2.5) pulls out test name, value, unit and range, with no judgement.
3. Rule engine: our own code parses the ranges, decides high/low/normal, computes urgency and flags likely OCR errors.
4. LLM explain: the model only sees the verified results and writes the summary, per-test meaning and doctor questions.
5. UI: Streamlit, with English and Bangla output.

Challenges

• In early tests the LLM labeled a high HbA1c value as "low" and wrote a wrong summary ("other parameters are normal"). That is dangerous in a medical setting, so we redesigned the pipeline: the LLM no longer judges values, a deterministic rule engine does.
• OCR sometimes misreads numbers and names (for example "41" instead of "4.1").
• Larger models were not available on our plan, so we worked with a smaller Qwen model and made the pipeline robust to its mistakes.

What we learned

LLMs are good at extracting and explaining, but should not be trusted alone with numbers in health use cases. Combining an LLM with simple deterministic checks makes the result much safer.

Real-world impact

ReportSimple helps patients and families understand results before and after a doctor visit, reduces panic and misinformation from searching numbers online, and lowers the language barrier for non-native English readers.

What's next

Age and sex specific ranges, more languages, tracking trends across reports, and a shareable summary for doctors.

Educational tool only. Not a diagnosis or a substitute for a doctor. Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for ReportSimple

Built With

• featherless-ai
• openai-api
• python
• qwen
• streamlit
• tesseract-ocr