SightBridge

Inspiration

Near my school, there was a place where blind people lived. Since childhood, I had the privilege of interacting with them regularly. Over time, I came to understand, at least in part, some of the real challenges they face every day. I also had experience helping educate blind and visually impaired people, which made those challenges even more real to me.

What stayed with me most was not only the difficulty of reading information, but also the emotional weight of depending on others for simple daily tasks. Something as ordinary as identifying the right medicine, checking the expiry date, understanding a receipt, recognizing a warning sign, or navigating in a new place can become stressful and risky when the world is designed visually.

That stayed with me for a long time.

SightBridge came from that experience and that feeling. I wanted to build something that could make life a little easier for blind and low-vision users, especially in those moments where fast, practical, spoken guidance can make a real difference.

I am building SightBridge for blind and low-vision people who face daily barriers because so much important information is presented visually.

Many tasks that sighted people do instantly still require assistance for visually impaired users, such as:

• reading a medicine label
• checking the dosage or expiry date
• understanding a receipt or bill
• identifying currency notes
• recognizing signs, symbols, or warning labels
• understanding what is in front of them in unfamiliar surroundings

What it does

SightBridge is built as a voice-first assistive system. A user can turn on the camera or upload an image, and the system interprets what is visible and responds with spoken guidance.

Core use cases

1. Explaining medicine

SightBridge can help answer questions like:

• What medicine is this?
• What is it for?
• How many mg is it?
• What is the dose?
• Is there an expiry date visible?

This is especially important because medicine-related mistakes can have serious consequences.

2. Explaining receipts and bills

It can read and summarize bills or receipts so users can understand charges, totals, dates, and other important details.

3. Identifying currency notes

It can help users identify paper notes, which supports independence in daily transactions.

4. Warning signs and directional guidance

It can detect warning-related symbols or visible signs and describe them in a practical way, helping users better understand their surroundings.

How I built it

I built SightBridge around a simple idea:

$$ \text{Visual Input} \rightarrow \text{AI Interpretation} \rightarrow \text{Clear Spoken Guidance} $$

The workflow is designed to be practical and accessible:

• The user captures an image through the camera or uploads a picture.
• The AI analyzes the visual content.
• It extracts useful details such as text, layout, object cues, labels, and contextual signals.
• The system turns that information into a clear, spoken explanation tailored for blind or low-vision users.

A major design choice was to focus on everyday tasks instead of abstract demonstrations. I wanted the tool to solve real problems people encounter regularly: medicines, receipts, currency, warning signs, and public information.

Challenges I ran into

1. Reliability and misinterpretation

The biggest challenge is that AI can be wrong. It may misread a medicine label, misunderstand a sign, or miss an important detail in a bill or form. In a project like this, that is not a minor issue.

Because of that, I had to think seriously about how the system should behave in uncertain situations. It should avoid sounding overly confident, clearly communicate uncertainty, and be especially careful in high-risk situations.

2. Safety in medical use cases

Medicine interpretation is one of the most valuable features, but also one of the most sensitive. A wrong dosage, expiry date, or label reading could be harmful.

So one challenge was designing the product mindset correctly: this tool should assist, not replace professional medical advice. In practice, that means adding safety messaging and encouraging verification with a pharmacist or doctor for important medical decisions.

3. Privacy

Users may scan personal documents, private bills, or public scenes involving other people. That creates privacy concerns.

A responsible version of the system should minimize data retention, avoid storing images by default, and make data handling transparent to the user.

4. Avoiding over-reliance

Another challenge is ensuring users do not treat the app as the sole source of truth in safety-critical situations. For example, it should not replace mobility aids, human support, or expert guidance. Framing the product correctly was therefore as important as building the features themselves.

Accomplishments that I’m proud of

One of the biggest accomplishments I am proud of is building a project with a real human purpose behind it. SightBridge was not created as just a technical demo. I built it to address a real accessibility challenge faced by blind and low-vision people in everyday life.

I am proud that, within a hackathon setting, I was able to design a tool that can help users interpret medicine labels, receipts, bills, currency notes, warning signs, and everyday surroundings through a voice-first AI experience. Instead of focusing only on text recognition, I aimed to make the system explain information in a way that is practical and meaningful for the user.

I am also proud that the project was shaped by personal experience and empathy. The inspiration came from years of interacting with blind people near my school and seeing some of the real barriers they face in daily life. Turning that long-held motivation into a working prototype feels deeply meaningful to me.

Another accomplishment is that I approached the project responsibly. I did not think only about what the AI can do, but also about what it should not do. I considered issues like medical safety, privacy, uncertainty, and over-reliance, and designed the idea as an assistive tool that supports people without taking decisions away from them.

Most importantly, I am proud that SightBridge reflects a simple but powerful goal: using AI to improve independence, dignity, and confidence for people who are often excluded by visually designed systems. For me, that is the most meaningful accomplishment of all.

What I learned

Building SightBridge taught me that accessibility is not just a technical problem. It is also a human-centered design problem.

I learned several important lessons:

1. Accuracy matters more in high-stakes contexts

A small mistake in reading a poster might be harmless, but a small mistake in reading a medicine label can be dangerous. That means assistive AI needs a very different standard of responsibility depending on the use case.

2. Context matters as much as text

Reading aloud raw text is not enough. Users often need meaning, structure, and explanation. Good accessibility tools must translate visual information into useful understanding, not just words.

3. Simplicity is powerful

For many users, especially in assistive settings, the best interface is often the simplest one. A voice-first interaction model can reduce friction and make the experience feel more natural.

4. Technology should preserve agency

The goal of accessibility technology should be to give people more control over their own lives, not less. That idea shaped the entire project.

What’s next for SightBridge

The next step for SightBridge is to move from a hackathon prototype into a more reliable real-world assistive tool.

My first priority would be to improve accuracy and safety, especially for high-stakes use cases like medicine labels, expiry dates, dosage instructions, warning signs, and bills. For these cases, the system should become better at handling uncertainty, highlighting important details, and avoiding overconfident responses.

I also want to make SightBridge more truly voice-first and hands-free, so that blind and low-vision users can interact with it more naturally in daily life. That means improving the spoken interaction flow, reducing friction in taking pictures or scanning scenes, and making the experience fast and simple.

Another important next step is real user feedback. I would want to test the product with blind and low-vision users, learn how they actually use it, understand what is helpful and what is frustrating, and redesign parts of the system based on their lived experience rather than assumptions.

In the future, I would also like to expand SightBridge’s capabilities beyond reading text. For example, it could provide better scene understanding, clearer directional guidance, stronger warning detection, and more context-aware explanations of surroundings.

At the same time, privacy and trust would remain central. I would want the system to minimize data retention, avoid storing sensitive images by default, and clearly communicate its limits so users stay in control.

Ultimately, my vision for SightBridge is to make it a dependable accessibility companion that helps blind and low-vision users navigate daily life with more independence, confidence, and dignity.

Built With

• anthropic-api
• camera/image-upload-pipeline
• css
• django
• html
• javascript
• ocr
• python
• text-to-speech