Manual exploration
QA engineers repeatedly navigate identical flows across builds.
LOCAL AI × REAL HARDWARE
Your AI QA Engineer,
Inside the Phone.
PocketQA autonomously explores Android apps on real smartphones, finds crashes, visual bugs and performance issues, generates reproducible reports, and verifies fixes by replaying the original journey.
✓ Local AI✓ Real smartphone testing✓ Autonomous exploration✓ Reproducible bugs✓ Automated verification
Built by Team OnCall Engineers
POCKETQAAGENT ACTIVE
Testing
com.demo.shop
Objective
Break the checkout flow
Current action
Changing shipping address
Progress78%
17actions6screens1issue
✓ Product discovered
✓ SAVE20 applied
● Changing shipping address
○ Awaiting UI response
LOCAL MODEL ACTIVE
The problem
Finding the bug is often harder than fixing it.
Developer ships buildQA manually testsBug appearsScreenshot takenBug ticket writtenDeveloper tries reproducingCannot reproduceBack to QAMore contextPatchQA tests again
Incomplete bug context
Checkout crashed rarely tells a developer what happened immediately before failure.
Reproduction takes time
Developers often spend significant time reproducing a problem before debugging starts.
Regression testing repeats everything
After every fix, someone must manually verify that the original issue disappeared.
Developers should spend their time fixing bugs, not rediscovering them.
The solution
From testing objective to verified fix.
Before
Traditional QA
- 01Write test
- 02Run test manually
- 03Find bug
- 04Take screenshots
- 05Write reproduction steps
- 06Assign developer
- 07Developer reproduces
- 08Fix
- 09QA retests
Manual work at almost every step
After
PocketQA
- 01Give PocketQA a goal
- 02AI explores app
- 03Bug detected
- 04Trace captured
- 05Reproduction generated
- 06Root cause suggested
- 07Developer fixes
- 08PocketQA replays flow
- 09FIX VERIFIED
One continuous automated testing loop
Interactive fake demo
Don't read about it.
Watch PocketQA break an app.
POCKETQAAGENT ACTIVE
Testing
com.demo.shop
Current action
Changing shipping address
SESSION PROGRESS78%
Local model is watching this screen.
POCKETQA AGENT
IDLE
Test checkout. Apply SAVE20, change the delivery address, and try to complete the purchase.
00:02Launching application
00:04Product feed detected
00:06Searching headphones
00:08Product selected
00:10Adding product to cart
00:12Opening checkout
00:14Testing coupon field
00:16SAVE20 accepted
00:18Modifying shipping address
00:20Confirming new address
00:22Unexpected application termination detected
The ROI
Less time reproducing.
More time shipping.
Illustrative example based on configurable assumptions.
Configurable assumptions
Adjust the inputs to model your workflow.
Recovery projection
Hours recovered / month~46 HRS
Estimated savings / month₹43,083
Estimated savings / year₹5,17,000
Manual QA hours reduced36.7 HRS
Illustrative estimate. Actual savings depend on team size, workflows, test complexity, and infrastructure.
One bug economics
The hidden cost of one bug.
Traditional
125 MINUTES
QA discovers bug15 min
QA reproduces15 min
QA writes report10 min
Developer reads ticket5 min
Developer reproduces25 min
Developer investigates20 min
Patch20 min
QA verifies15 min
PocketQA assisted
40 MINUTES
PocketQA discoversAutomated
Trace generatedAutomated
Developer reviews trace8 min
Developer investigates12 min
Patch20 min
PocketQA verifiesAutomated
Example human time reduction: 85 minutes per bug
Illustrative workflow example, not a measured production benchmark.
Device telemetry
The emulator cannot feel the phone getting hot.
PocketQA correlates failures with actual device behavior on real Android hardware.
frame drops·memory spikes·thermal behavior·network instability
CPU
43%
RAM
742 MB
FPS
58
TEMP
39.4°C
BATTERY
3.2% / hr
NETWORK
82 ms
Local AI
Your application's screen does not need to leave the phone.
LOCAL AI STATUS
Model
On-device VLM
Inference
LOCAL
Cloud requests
0
Screen data uploaded
0 bytes
Status
ACTIVE
PocketQA is designed around local and open-source models for screen understanding and reasoning. Sensitive application context can remain on the testing device.
ScreenshotOn-device vision modelUI understandingAndroid interaction
Architecture
Phone first by design.
PHONE
1Target Android app
2Screen capture
3Touch / interaction layer
4Voice context
5Hardware telemetry
6Local VLM / LLM
7PocketQA agent
8Structured test trace
DEVELOPER ENVIRONMENT
1Codebase
2Code retrieval
3Root cause analysis
4Patch / regression test
5Send new build
6Replay original journey
7FIX VERIFIED
Office Kit can bridge phone and developer environments during the hackathon workflow.
Capabilities
More than scripted UI automation.
Goal driven testing
Describe what should work instead of programming every interaction.
Autonomous exploration
AI determines which screens and paths should be tested.
Visual understanding
Understands interfaces semantically beyond selectors.
Real hardware telemetry
Observes actual performance, thermal, and resource behavior.
Reproducible traces
Preserves the journey that caused a failure.
AI root cause assistance
Connects runtime behavior with relevant source context.
Regression generation
Turns discovered failures into future tests.
Automated verification
Replays the original failing journey after a fix.
Previous work
Built on things we already know how to ship.
PocketQA combines ideas our team has explored across local AI, autonomous agents, interaction replay, and production software.
Local Multimodal AI
Hey Dot
Fully local voice-first AI assistant that understands the user’s screen using local vision-language models.
View GitHubLocal AIVision understandingVoiceScreen context
Interaction Recording and Replay
RecordScrape
Visual browser automation platform that records interactions and converts them into repeatable workflows.
View GitHubInteraction captureReplayAutomationState tracking
Autonomous Voice Agents
ComplaintBuddy AI
AI agent that interacts with customer service systems on behalf of users.
View GitHubAgentsAutonomous decisionsReal-world workflowsAI orchestration
The builders
Team OnCall Engineers
We build AI systems that interact with the real world, not just chat interfaces.
Builder 01
Shubham Vishwakarma
University of London · Computer Science, AI/ML
Studying AI and Machine Learning at the University of London on a performance scholarship. Shubham has programmed since age 10 and sold his first product at 12. Since then, he has contracted independently and worked as a software engineer and founding engineer for several companies in the US, Dubai, and the UK.
AI systemsbackend architecturelocal LLMsautomationRAGdeveloper infrastructure
Builder 02
Srushti J Sundar
Saptagiri NPS College
Content creator who has collaborated with 40+ brands, bringing a strong audience and product storytelling perspective to the team.
FOCUS
Product engineeringAIApplication developmentProduct experience
Hackathon achievements
Built under pressure.
Repeatedly.
6×
National Hackathon Winners
A track record of turning ambitious ideas into working products and strong live demos.
2026
Top 10
HackBLR, Bengaluru
2025
Winner
CodeClash 2.0 at Google
2025
Top 10
Code for Bharat at Microsoft · 3000+ teams
2025
3rd Prize and Top 10
Code for WIE 3.0, MSIT Delhi · 300+ teams
2024
Top 15
Code the Cubicle 3.0 at Mastercard · 3500+ teams
2024
2nd Place
Live the Code 3.0 · GeeksForGeeks
2024
Top 10 and Special Mention
InnovateX · Delhi Technological University
2024
Best Rookie Hack
Tech Ingenium 5.0 · Ahmedabad University
Why us
Why Team OnCall Engineers?
Proven hackathon execution
We have repeatedly built, demonstrated, and placed at national-level hackathons.
Existing technical foundation
Interaction replay, local VLMs, AI agents, and automation are areas we have already built in.
Product first
Our goal is a working product with a compelling end-to-end demo, not disconnected AI features.
Built for this hardware
PocketQA makes the smartphone part of the intelligence and testing infrastructure.