Priyanshu Pandey - Android Developer

Overview

Android engineer at Senior level specializing in native Kotlin apps built with Jetpack Compose and platform integrations. Strongest proven skill is platform integration and device-feature centric app design demonstrated by ChallengeActivity.kt sensor and input handling together with AlarmScheduler.kt persistence and scheduling. Public code shows little evidence of CI/release automation, systematic performance benchmarking, or an offline sync/conflict-resolution engine.

Phone

Timeline

United College of Engineering and Research

Master's Degree · Computer Science & Engineering (Specialization in AI & ML)

2023–2027 Prayagraj, Uttar Pradesh

Software Engineering & AI/ML Intern · Junior

HCL Technologies · Internship

Jun 2024 to Aug 2024 2 Months Partially remote

Remote internship focused on software engineering and AI/ML work. Supported development activities for internal initiatives under an internship program. Contributed to engineering tasks while working in a remote team environment.

Generative AI & Software Intern · Junior

IBM · Internship

Jan 2024 to Mar 2024 2 Months Partially remote

Remote internship focused on generative AI and software engineering. Assisted with development and experimentation related to generative AI workflows. Worked on supporting engineering tasks for AI-enabled features in a remote setting.

Projects

Smart Fog Vision: AI Dehazing & Driver Hazard Alert System

Jan 2024 to Present

Computer vision system for foggy driving using an edge/on-device workflow. Implemented dehazing using Dark Channel Prior with refinement steps, and built detection/segmentation with YOLOv8n and TinyUNet. Added fog-density estimation for visibility and hazard alerting, using ONNX Runtime and OpenCV for the processing pipeline.

TaskIOn Real-Time On-Demand Task & Service Marketplace

Jan 2024 to Present

Android on-demand task and service marketplace inspired by TaskRabbit. Added real-time task telemetry and bidirectional state sync using WebSockets and Redis Pub/Sub, with interactive Kanban task boards. Implemented offline-first caching with Room/SQLite, proximity matching using GPS, and alert notifications via FCM for client and tasker updates.

Avenue & Co Multi-Vendor Marketplace Mobile App

Jan 2024 to Present

Android marketplace app built with Kotlin and Jetpack Compose using MVVM and Clean Architecture. Implemented REST API integrations for multi-merchant split orders and secure checkout. Added offline-first storage and persistence with Room/SQLite, plus push notifications with FCM and Redis-backed locking for atomic inventory reservation.

Technical skills

Languages

9

Python

Kotlin

Node JS

JavaScript

Java

SQL

Dart

Swift

Objective-C

Node JS

3

Axios

Express

Sharp

Kotlin

4

Kotlin Coroutines

kotlinx.serialization

OkHttp

Retrofit

Mobile

13

Jetpack Compose

Android SDK

Material3

Firebase

Flutter

React Native

SwiftUI

Core ML

Room

WorkManager

ML Kit

ViewModel

Android

AI/ML

11

PyTorch

Datasets

YOLO

NumPy

ONNX Runtime

LLM

OpenCV

Machine Learning

TensorFlow

LiteRT

Deep Learning

Frontend

5

React.js

Tailwind CSS

React Router

Redux

Vite

Other

28

Marshmallow

Spring Boot

FastAPI

MySQL

Redis

FAISS

GitHub

SQLite

Computer Vision

PostCSS

Docker

Git

Rest API

Postman

Material Design

ONNX

CI/CD

RAG

autoprefixer

MVVM

Image Segmentation

WebSockets

Clean Architecture

State Management

Edge AI

MVC

Offline-First

Hallucination

Senior AI/ML Engineer Confidence: High ML Engineer

Senior ML engineer focused on computer vision for adverse weather driving conditions with strong practical end-to-end training and deployment work. The strongest proven skill is building complete CV training and deployment pipelines including streaming dataset ingestion, custom model implementations and browser-ready ONNX exports as shown in train_streaming_fog.py, train_road_seg.py and train_fog_classifier.py. Public code does not show production-grade experiment tracking, unit tests, distributed training or formal monitoring and CI/CD integration.

Model Architecture & Training

5/10

How well models are designed and trained

Solid custom model work and end-to-end training loops across multiple tasks (dehazing, segmentation, classification). Shows architecture choices, loss composition, schedulers and training routines but no evidence of large-scale distributed training or formal ablation studies.

Evidence

training/train_streaming_fog.py:class AODNet and train_dehaze training loop

training/train_road_seg.py:class TinyUNet and training loop

training/train_fog_classifier.py:MobileNetV2 head replacement and training loop

Data Pipeline & Feature Engineering

6/10

How data is prepared for models

Well-developed data ingestion and augmentation pipelines including streaming large datasets, synthetic fog augmentation and dataset conversion for YOLO. Demonstrates stratified splits, robust parsing of multiple dataset layouts and on-the-fly streaming to avoid local download.

Evidence

training/train_streaming_fog.py:class StreamingFogDataset (HF streaming, streaming=True)

training/train_road_seg_custom.py:add_synthetic_fog and CustomRoadDataset

training/train_lowlight_detector.py:prepare_dataset and write_split (YOLO dataset construction)

Experimentation & Evaluation

4/10

How results are measured and tested

Reasonable experiment and evaluation practices: per-epoch validation, metric computation, model card and history saving, and simple PSNR/IoU/F1 reporting. Lacks experiment tracking integration (W&B/MLflow) and formalized reproducible experiment harness.

Evidence

training/train_fog_classifier.py:compute_metrics, evaluate and history.json exports

training/train_road_seg.py:evaluate function and metrics_road_seg_val.json saving

training/train_streaming_fog.py:validation PSNR logging and history_dehazer.json

MLOps & Deployment

5/10

How models are shipped to production

Clear deployment focus with ONNX export for browser inference, export-only modes, model cards and exporting pipeline packaging. Shows practical attention to model portability, but no evidence of CI/CD, monitoring or automated serving orchestration.

Evidence

training/train_fog_classifier.py:torch.onnx.export block and ONNX verification

training/train_road_seg.py:ONNX export and export-only mode

training/train_streaming_fog.py:export_dehazer_onnx and export flags

Computational Efficiency

5/10

How efficiently computing resources are used

Multiple pragmatic efficiency choices: small/light AODNet for on-device inference, freezing backbone layers, gradient clipping, and streaming to avoid disk IO. No low-level GPU profiling, quantization experiments with measured before/after metrics, or Triton/CUDA kernel work is present.

Evidence

training/train_streaming_fog.py:AODNet described as ultra-lightweight and suited for on-device inference

training/train_fog_classifier.py:freezing features[:14] to speed training and reduce compute

training/train_streaming_fog.py:streaming dataset approach to avoid downloading large datasets

Research Depth & Innovation

3/10

Depth of research and new ideas

Implements paper-model reproductions and domain-specific augmentations (AODNet, TinyUNet, synthetic fog) but does not present novel research contributions, formal ablation studies or peer-reviewed-level replication experiments.

Evidence

training/train_streaming_fog.py:AODNet implementation (paper reimplementation)

training/train_road_seg.py:TinyUNet model code

training/train_road_seg_custom.py:add_synthetic_fog function

Expertise

Computer Vision & Image Analysis· Middle

MLOps & Model Lifecycle· Middle

Industries

Transportation & Logistics· Senior

Technologies

Deep Learning

Python· since 2024 · Senior

SQL

MySQL

Redis

OpenCV

FAISS

YOLO

WebSockets

Computer Vision

ONNX

Datasets

CI/CD

TensorFlow· since 2026

NumPy

Git

SQLite

PyTorch· since 2024

Docker

LLM

RAG

Marshmallow

Hallucination

Image Segmentation

GitHub

Edge AI

LiteRT

ONNX Runtime

Machine Learning· since 2026

Systems· mentioned only

Vision· mentioned only

Recommendations

Develop lightweight, real-time ADAS pipelines that convert research models into browser/edge-ready ONNX artifacts and client-side inference.
Build robust dataset ingestion and augmentation tooling for large vision corpora, including streaming, synthetic-weather augmentation and automated split/metadata indexing.
Integrate experiment tracking and reproducibility (W&B or MLflow) plus CI/CD packaging for ONNX model releases and regression monitoring.
Work on domain adaptation and quantization flows with before/after measured metrics to optimize on-device latency and memory.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Smart_Fog_Vision_Sytem Smart-Fog-Vision-System Priyyanxhu

Senior Mobile Developer Confidence: High Android Engineer

Android engineer at Senior level specializing in native Kotlin apps built with Jetpack Compose and platform integrations. Strongest proven skill is platform integration and device-feature centric app design demonstrated by ChallengeActivity.kt sensor and input handling together with AlarmScheduler.kt persistence and scheduling. Public code shows little evidence of CI/release automation, systematic performance benchmarking, or an offline sync/conflict-resolution engine.

Platform Native Mastery

6/10

Knowing the mobile platform

Good platform-native mastery: explicit handling of lockscreen, foreground service workflow, TTS lifecycle, sensor registration/unregistration, SharedPreferences persistence and boot-time migration; lifecycle concerns handled in activities and DisposableEffect hooks in Compose.

Evidence

Priyyanxhu/Finch: app/src/main/java/com/smartalarm/ChallengeActivity.kt - setShowWhenLocked/setTurnScreenOn, enforceMaxVolume, onBackPressedDispatcher and dispatchKeyEvent handling Priyyanxhu/Finch: app/src/main/java/com/smartalarm/MainActivity.kt - TextToSpeech lifecycle (DisposableEffect), checkExactAlarmPermission and notification permission launcher Priyyanxhu/Finch: app/src/main/java/com/smartalarm/AlarmScheduler.kt - loadAll/saveAll persistence, legacy migration and trigger scheduling logic

Mobile UI/UX & Responsiveness

5/10

Smooth mobile experience

Strong Jetpack Compose usage and Material 3 application with thoughtful UI patterns and state handling; good use of rememberSaveable, Lazy lists and adaptive components but no evidence of multi-posture/foldable or extensive accessibility/RTL tuning.

Evidence

Priyyanxhu/Finch: app/src/main/java/com/smartalarm/MainActivity.kt - SmartAlarmAppContent, AlarmsTabContent and AlarmEditorDialog Compose implementations Priyyanxhu/Finch: app/src/main/java/com/smartalarm/ChallengeActivity.kt - many composable tasks (MathTask, HoldButton, SliderTask, etc.) showing UI/UX work

Performance & Battery

4/10

Speed and battery use

Platform-aware performance and battery decisions are present (foreground service for alarm, use of sensor sampling rates and explicit unregister), but there are no measured performance reports, baseline profiles, or documented jank/memory measurements.

Evidence

Priyyanxhu/Finch: app/src/main/java/com/smartalarm/ChallengeActivity.kt - rememberAccel/rememberLux use SensorManager with SENSOR_DELAY_GAME / SENSOR_DELAY_NORMAL and explicit unregister in DisposableEffect Priyyanxhu/Finch: app/src/main/java/com/smartalarm/MainActivity.kt - uses ContextCompat.startForegroundService for alarm testing and stops/releases MediaPlayer resources in stopAudio

Offline & Data Sync

2/10

Working offline and syncing

Basic local persistence and migration implemented with SharedPreferences; no evidence of an offline-first sync engine, queued idempotent retries, migration history beyond a simple legacy migration, or conflict resolution.

Evidence

Priyyanxhu/Finch: app/src/main/java/com/smartalarm/AlarmScheduler.kt - loadAll/saveAll using SharedPreferences and KEY_ALARMS_JSON with legacy key migration

Device Integration

6/10

Using device features

Multiple device integrations and permission flows are implemented: sensors, camera intent, speech recognizer, TTS, step counter fallback, and notification/exact-alarm permission prompting with user feedback.

Evidence

Priyyanxhu/Finch: app/src/main/java/com/smartalarm/ChallengeActivity.kt - VoiceTask uses RecognizerIntent, PhotoTask uses TakePicturePreview launcher, sensor-based tasks use accelerometer/step counter and fallback logic Priyyanxhu/Finch: app/src/main/java/com/smartalarm/MainActivity.kt - notification permission request (POST_NOTIFICATIONS) and checkExactAlarmPermission requesting Settings.ACTION_REQUEST_SCHEDULE_EXACT_ALARM

Release & App Lifecycle

1/10

Building and publishing apps

No clear release engineering artifacts found; no CI, fastlane, signing configs, ProGuard/R8 rules, crash reporting wiring, or store release metadata are present in the analyzed human-authored files.

Expertise

Android· Senior

Mobile UI/UX· Middle

Industries

Commerce· Middle

Transportation & Logistics· Senior

Financial Services· Middle

Technologies

Android

JavaScript

Java

Kotlin· since 2024 · Senior

Dart

Swift

Jetpack Compose· since 2024

Objective-C

SwiftUI

Android SDK

Core ML

Retrofit

MVVM

Flutter

React Native

Room

Kotlin Coroutines

Firebase

Clean Architecture

kotlinx.serialization

OkHttp

MVC

State Management

Offline-First

WorkManager

ML Kit· since 2026

Material Design

ViewModel

Material3

Clean Architecture· mentioned only

Room· mentioned only

Recommendations

Lead development of Android-native features that require close platform integration: alarms, sensors, camera, TTS and foreground service workflows.
Implement end-to-end alarm reliability improvements: complete AlarmForegroundService scheduling code audit, add alarm tests and integration tests for BootReceiver/reschedule flows.
Add CI/release automation and baseline performance measurements (startup, memory, jank) and document optimizations and trade-offs.
Build a small offline-sync engine or queue for user operations if the product needs cross-device consistency or cloud sync with idempotent retries.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Finch TaskIon UrbanBreeze

Middle Frontend Developer Confidence: Medium Fullstack

Fullstack Engineer (senior-level baseline) focused on building end-to-end image matching and age-progression APIs. The strongest proven skill is backend image-and-matching system design, demonstrated by the matching engine and scoring logic in backend/utils/matching.js and the end-to-end flow in backend/server.js. Public code does not show a mature frontend component library, formal test coverage or CI-driven quality gates, and there is limited evidence of measured performance or accessibility work.

UI Component Architecture

2/10

How interface parts are built

Minimal frontend component architecture evidence; project contains scaffolding and Tailwind theming but no custom React component library, composition patterns or component lifecycle logic.

Evidence

frontend/index.html: root mount and fonts but no React components

frontend/tailwind.config.js: custom theme/colors and fonts (visual tokens present)

Responsive & Cross-browser

2/10

Works on all screens and browsers

Basic responsive tooling present (Tailwind, viewport meta and fonts) but no advanced responsive strategies, container queries, RTL or explicit cross-browser feature detection shown.

Evidence

frontend/index.html: viewport meta and font preconnect

frontend/tailwind.config.js: breakpoints and theme extensions

Performance Optimization

3/10

Speed of the interface

Some backend performance awareness (SQLite PRAGMA, indexes, WAL) and sensible request limits, but no measured LCP/INP artifacts, bundle analysis or advanced client-side optimizations.

Evidence

backend/db.js: PRAGMA journal_mode = WAL; CREATE INDEX statements

backend/server.js: multer limits and use of memoryStorage for controlled upload sizes

backend/utils/matching.js: lightweight deterministic generateEmbedding to avoid heavy ML dependencies

Accessibility & Semantics

1/10

Usable for everyone

Almost no accessibility or semantic UI work is present beyond basic HTML lang and meta; no ARIA, focus management, keyboard handling or CI a11y tooling visible.

Evidence

frontend/index.html: lang attribute and basic head meta only

State Management & Data Flow

4/10

Managing data in the app

Clear server-state discipline and data flows with validation, migrations, careful photo ordering and edge-case handling (prevent deleting last photo); request-level checks exist but there is no advanced request cancellation, optimistic updates or formal state machines.

Evidence

backend/server.js: route validations, file handling, upload limits and age-progression endpoints (/api/age-progress, /api/age-progress-all)

backend/db.js: create/insert/update functions, deleteLostPhoto validation preventing last-photo deletion and re-normalizing sort_order

backend/utils/matching.js: computeMatchScore handling multi-photo embeddings and breakdown of matching factors

UX & Visual Polish

2/10

Look and feel quality

UX-focused endpoints and features exist for previews and bulk age progression, but frontend UX artifacts (skeletons, keyboard flow, undo patterns or measured perceived-performance improvements) are not present.

Evidence

backend/server.js: /api/age-progress (preview and saveAsNew) and /api/age-progress-all endpoints

backend/utils/ageProgression.js: generateAgeProgressedImage returns imageBase64, description and confidence for preview UX

Expertise

React· Junior

Frontend Architecture & Build Tools· Junior

Industries

Artificial Intelligence· Middle

Government· Middle

Technologies

Node JS· Middle

Redux

Tailwind CSS

Express

React.js

Vite

Axios

PostCSS

Sharp

React Router

autoprefixer

Recommendations

Lead development of backend APIs that perform image processing, embedding generation and matching for prototypes or NGO/police tooling.
Implement and own the integration of real ML model endpoints (insightface/replicate) and replace the simulated embedding/progression functions with model-backed services.
Build the frontend consumer experience for matching previews and verification flows, adding accessible custom components and comprehensive UX states.
Add automated tests (unit and integration), CI pipelines and basic performance benchmarking to harden the system for production.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Chrono

Job preferences

Work permits: India
Citizenship: India
Salary: from $10k per year
Employment type: Full-Time