Inspiration
I have always wanted to use technology to do something genuinely good for society. We use technology every day to make work easier and complete routine tasks, but it can also improve someone’s quality of life.
At the same time, electronic waste is growing rapidly. People replace phones frequently, while disposable, donated, and unused devices accumulate. Many still have a working camera, processor, and internet connection.
That inspired Akshrava: if a recycled phone could help change someone’s life, giving it a second purpose would be worth doing.
What it does
Akshrava transforms affordable, recycled, donated, and repurposed Android phones—and potentially iOS devices—into assistive-vision companions for people with visual impairments.
Using the phone camera and cloud-based computer vision, it detects supported people, vehicles, and obstacles, then provides conservative spoken and haptic cues such as:
- “Person ahead”
- “Vehicle nearby, left”
- “Obstacle ahead”
Akshrava is designed to complement—not replace—a cane, guide, or mobility training.
How we built it
The Android application uses CameraX, adaptive frame compression, encrypted device-token storage, WebSocket streaming, Android text-to-speech, haptic feedback, and reconnect handling.
The backend uses FastAPI, YOLO-based remote inference workers, PostgreSQL through Cloud SQL, Redis protections, Terraform-managed Google Cloud infrastructure, calibration profiles, alert policies, and operational telemetry.
We designed the system around safety boundaries: stale detections are suppressed, unverified calibration fails closed, and sensitive tokens, images, locations, and device identities are not written to logs.
Challenges we ran into
Supporting older phones was one of our biggest challenges. Recycled devices can have limited memory, slower processors, different cameras, aggressive battery management, and unreliable network connections.
We also had to handle:
- stale or delayed inference results;
- secure device provisioning;
- dynamic image scaling;
- distributed replay protection;
- calibration accuracy;
- foreground-service restrictions;
- Android-version compatibility;
- clear differentiation between “nothing detected,” “detector unavailable,” and “network failure.”
Accomplishments that we’re proud of
We built an end-to-end system that connects an older Android phone to cloud vision infrastructure and provides real-time spoken and haptic awareness cues.
We are especially proud of:
- reusing hardware that might otherwise become electronic waste;
- supporting multiple generations of Android devices;
- adding secure, device-bound authentication;
- creating conservative safety and freshness policies;
- preserving privacy by avoiding normal image storage;
- building diagnostics that help identify failures without exposing sensitive data;
- testing the complete phone-to-cloud protocol rather than only testing the detector.
What we learned
We learned that assistive technology is a systems problem, not just a machine-learning problem. Detection accuracy matters, but so do latency, battery life, connectivity, privacy, calibration, accessibility, and failure behavior.
We also learned that a late or overconfident alert can be worse than no alert. For that reason, Akshrava never speaks results beyond its freshness limit and never claims distance or urgency without verified calibration evidence.
What’s next for Akshrava
Next, we want to:
- deploying technology widely so we can help as many people as possible.
- complete longer supervised field testing;
- improve performance through GPU inference and regional scaling;
- expand support for more recycled Android devices;
- investigate a future iOS client;
- improve calibration workflows;
- add more carefully validated object classes;
- strengthen accessibility and multilingual support;
- reduce battery and bandwidth usage further;
- build partnerships with accessibility organizations, refurbishers, and community programs.
Our long-term goal is to make assistive awareness more affordable, sustainable, and widely available by turning existing devices into tools that can make a meaningful difference.
Log in or sign up for Devpost to join the conversation.