Open to opportunities

Kunwar Partap

@kunwarpartap

Message

I build offline on-device AI Android apps for mental health and accessibility.

India

Message

What I'm looking for

I’m looking for student researcher or early engineering roles where I can build privacy-first, offline AI systems—especially Android apps for on-device NLP and accessibility—plus real-time data pipelines and AWS-based monitoring when needed.

I’m a B.Tech Computer Science Engineering student who builds privacy-first, offline AI experiences—grounded in real user needs and designed to run entirely on-device. My work focuses on on-device AI, Computer Vision, NLU, and mobile development, with an emphasis on reliability and low-latency interactions.

In my B.Tech final year project, TETRA — Offline AI Mental Health Companion, I engineered a fully offline Android AI companion that runs Gemma 3 270M entirely on-device via MediaPipe—“no internet, no data leaves the phone.” I implemented deterministic crisis detection scanning for English/Hindi keywords before any AI involvement, routing to VANDREVALA & KIRAN helplines with 100% reliability, and built a multi-screen app architecture with voice-driven offline sessions using Android SpeechRecognizer (hi-IN, en-IN) and Android TTS.

I’m also developing ATOM — Offline Voice-Controlled Android Accessibility System, integrating Vosk speech recognition directly with the Android microphone pipeline to achieve zero latency and using an intent parsing approach with 95% accuracy. Beyond mobile, I built systems like a real-time event processing pipeline with Flask + MongoDB (500+ GitHub webhook events) and a Computer Vision Gesture Recognition System in Python with OpenCV and MediaPipe—delivering 95% typing accuracy at 30 FPS—plus cloud infrastructure monitoring on AWS (EC2/CloudWatch/SNS/IAM) to improve uptime by 25%.

Experience

Work history, roles, and key accomplishments

Current

TETRA Offline Mental Health

Current

TETRA — Offline AI Companion

Jan 2026 - Present (4 months)

Built a fully offline Android mental health companion using Gemma 3 270M on-device via MediaPipe, with no internet/data leaving the phone. Implemented deterministic crisis keyword detection with 100% reliability routing users to VANDREVALA and KIRAN helplines, plus offline voice-driven CBT/DBT/ACT sessions and weekly PDF journal exports.

Mediapipe Android Development Java MediaPipe Deterministic Safety Routing

Current

ATOM Offline Voice Accessibility

Current

ATOM — Offline Voice Accessibility

Jan 2026 - Present (4 months)

Developed an offline, low-latency voice-controlled Android accessibility system by integrating the Vosk speech engine directly into the microphone pipeline. Created a 3-layer architecture with an intent parser achieving 95% accuracy and an Instagram UI agent for coordinate-based interactions, with planned on-device Gemma 2B NLU.

Kotlin Jetpack Compose Vosk Speech Recognition Android Accessibility Services

Real-Time Event Processing

Real-Time Event Processing System

Jan 2026 - Present (4 months)

Delivered a real-time monitoring system that processed 500+ GitHub webhook events with zero data loss using a Flask REST API. Reduced storage by 60% via optimized MongoDB schema design and built a dashboard with 15-second auto-refresh for immediate event visibility.

Python Flask MongoDB REST APIs Webhooks JavaScript Data Normalization Monitoring

Gesture Recognition System

Computer Vision Gesture Recognition

Apr 2024 - Present (2 years 1 month)

Built a virtual keyboard using OpenCV and MediaPipe with an 85% gesture recognition accuracy and efficient hand tracking. Achieved 95% typing accuracy at 30 FPS using MediaPipe landmark detection.

Mediapipe Python OpenCV MediaPipe Computer Vision Machine Learning Hand Landmark Detection Virtual Keyboard

AWS Cloud Infrastructure Monitoring

AWS Cloud Infrastructure Monitoring System

Feb 2024 - Present (2 years 3 months)

Improved EC2 system uptime by 25% by automating CloudWatch monitoring with proactive alerting for CPU threshold violations. Implemented secure AWS architecture using IAM roles, security groups, and SNS email alerts for real-time incident response.

AWS EC2 AWS Cloudwatch AWS IAM Monitoring And Alerting Security Architecture Incident Response Linux

Education

Degrees, certifications, and relevant coursework

Chandigarh Group of Colleges

Bachelor of Technology, Computer Science Engineering

Activities and societies: Relevant coursework: Data Structures, Algorithms, Cloud Computing, Software Engineering, DBMS, Operating Systems, ML.

Pursuing a B.Tech in Computer Science Engineering at Chandigarh Group of Colleges, expected to graduate in June 2026. Relevant coursework includes Data Structures, Algorithms, Cloud Computing, Software Engineering, DBMS, Operating Systems, and ML.