FULLSTACK DEVOPS ENGINEER

SRE / INFRASTRUCTURE RELIABILITY / AUTOMATION

Linux · Docker · Kubernetes · Helm · Kafka · PostgreSQL · Prometheus · Grafana · Python · Bash · PowerShell · GitHub Actions

SYSTEM INITIALIZING...Loading Android SDK...Loading Flutter runtime...Compiling Compose...Loading AI context...
Available for software engineering opportunities

Fullstack DevOps Engineer / Mobile • Web • Backend • CI/CD • Infrastructure & Observability

Fullstack DevOps Engineer,
Mobile • Web • Backend • CI/CD • Infrastructure & Observability.

Fullstack DevOps Engineer with 6+ years of professional software development experience, specializing in native Android engineering and building solutions across mobile, frontend, backend, automation, containerization, CI/CD, and infrastructure observability.

AndroidKotlinJetpack ComposeReact NativeNode.jsAI / ML

SamePerson.DifferentPerspective.

Scroll to explore
01 / Featured engineering project
Ongoing ProjectIndependent Engineering Project

SENTINELOPS

Infrastructure Reliability & Automated Recovery Platform

Building an infrastructure reliability platform focused on service observability, incident detection, automated diagnostics, and controlled recovery workflows.

PROJECT ARCHITECTURE / PLANNED STACK
LinuxDockerKubernetesHelmKafkaPostgreSQLPrometheusGrafanaPythonBashPowerShellGitHub Actions
CORE OBJECTIVES
  • Infrastructure and service health monitoring
  • Incident detection and event-driven processing
  • Centralized metrics and operational dashboards
  • Automated diagnostics and controlled recovery
  • Container orchestration and deployment automation
PROPOSED ARCHITECTUREPlanned workflow
01Infrastructure / ContainersLinux · Docker · Kubernetes
02Health Checks & MetricsService signals and runtime health
03Prometheus / MonitoringCentralized observability layer
04Incident DetectionIdentify and classify events
05Kafka Event ProcessingEvent-driven processing path
06Diagnostics & Recovery AutomationControlled actions and verification
07PostgreSQL Incident History / Grafana DashboardsTraceable records and operational visibility
01 / About

Fullstack DevOps Engineer,
architecture first.

Fullstack DevOps Engineer with 6+ years of professional software development experience, specializing in native Android engineering and building solutions across mobile, frontend, backend, automation, containerization, CI/CD, and infrastructure observability.

My core strength is production Android engineering: Kotlin, Jetpack Compose, performance work, modular architecture, and long-running codebase modernization. I expanded into React Native and web technologies while preserving product behavior and clear system boundaries.

I build AI features as real product systems: authenticated flows, backend contracts, credits, payments, Supabase Edge Functions, OpenAI/LLM integration, and professional-context experiences like Digital Me.

Download resume
Mobile EngineeringAndroid Native / Kotlin
Compose / React Native
Backend & APIsNode.js / TypeScript
REST / Supabase
Web & DesignHTML5 / CSS3
Figma / Responsive UI
Cloud & DevOpsAWS / Firebase
Cloudflare / Vercel
AI & AutomationCodex / Claude
LLMs / Computer Vision
Based InGuwahati
Assam, India
02 / Digital Me

Don't just read my portfolio.
Talk to it.

Digital Me is an interactive alternative to a static resume.

Try asking

0Published apps
0Production projects
0Years experience
0Compose screens
60%Legacy MVP migration
03 / Featured projects

Production apps,
tools, and systems.

A recruiter-first view of Android, Flutter, AI, SDK, architecture tooling, and low-latency streaming work.

ARCHGUARD
02 / Open-source Gradle plugin

ArchGuard

Value: Executable architecture rules for Android teams that want feature-first validation and Clean Architecture enforcement.

Challenge: Turn architecture conventions into repeatable Gradle checks and readable HTML reports instead of relying only on code review memory.

KotlinGradle PluginArchitecture rulesHTML reportsDeveloper toolingMCP-ready tooling
Explore ArchGuard
BIOMETRIC SDK
04 / Reusable Android SDK

Biometric SDK

Value: A reusable Android SDK that wraps biometric authentication behind a cleaner integration surface.

Challenge: Package lifecycle-aware BiometricPrompt behavior into developer-friendly APIs that can be reused across apps.

KotlinAndroid SDKBiometricPromptLifecycle-awareReusable APIs
Explore Biometric SDK
04 / Adaptive Exercise Recognition

From Rule-Based Detection
to Personalized ML.

A Vantage Fit case study in production computer vision, anti-cheat design, TensorFlow Lite inference, and iterative engineering.

Problem / Iteration / Production ownership

Detect real exercise movement with a mobile camera, then keep improving after real users found ways to break it.

The requirement was to detect exercises such as squats using the mobile camera. I used an existing pose/skeleton detection API to obtain labeled body landmarks; I did not build the pose detector itself. The engineering work was the recognition system layered on top: movement tracking, validation gates, anti-cheat checks, ML classification, rep counting, and personalization.

Build -> Observe -> Users Break It -> Understand Why -> Improve -> Repeat
Camera Frame->Pose Landmarks->Joint Movement Tracking->Waveform / Cycle Detection->Confidence Threshold->Squat Count
WHEN USERS STARTED CHEATING

Recorded sessions became engineering feedback.

REAL-WORLD TESTING

During real-world testing, I used background session recording to understand how users were actually interacting with the feature. Normal QA could validate expected paths, but it could not predict every creative way someone would try to game the detector.

Some of the funniest user behavior produced some of the most valuable engineering lessons. Users often believed they were defeating the camera algorithm itself. In reality, each attempt simply exposed another assumption in the system that could be measured, understood, and hardened.

Unexpected user behaviorFalse assumptions in the algorithmNew cheating strategiesEnvironmental reliability issues
Iteration 01

Rule-Based Squat Detection

The first implementation tracked relevant joints over time and modeled the squat as a movement cycle similar to a sine waveform. If the observed cycle matched the expected squat pattern at roughly the original 90%+ confidence threshold, the system counted one squat.

Exploit Found

Lying Down

From recorded session data, I saw users lying down and repeatedly moving their legs. Because the first detector focused heavily on movement cycles, that behavior could sometimes resemble a valid squat pattern. I added standing-posture validation and skeletal landmark orientation checks, so a valid cycle only counted when the body posture was also valid.

Valid Squat Cycle+Standing Posture=Valid Squat
Fix

Phone Orientation

The next loophole was phone orientation. Users discovered that changing how the phone was positioned could interfere with assumptions made by the camera coordinate system. I solved it with Android device-orientation APIs, requiring the session to start only when the phone was in an expected valid orientation.

Instead of making the vision algorithm unnecessarily complicated, use device information the operating system already provides.
Input Gate

Poor Lighting and Landmark Jitter

Poor lighting made pose landmarks unstable, creating motion noise and occasional false detections. I introduced a lighting quality gate so bad input was rejected before reaching the expensive recognition pipeline.

Garbage input should be rejected before reaching the expensive recognition pipeline.
UPDATED VALIDATION PIPELINE

Reject weak signals early.

Camera->Lighting Validation->Device Orientation Validation->Pose Detection->Standing Validation->Exercise Detection
ML Upgrade

Moving Beyond Hand-Written Rules

As the feature matured, the waveform approach became harder to generalize. I replaced the sine-wave classifier with a TensorFlow Lite model trained to recognize multiple exercise patterns, initially including squats, push-ups, and jumping jacks.

The pose system still provided body movement data, but classification moved from fixed rules to learned movement recognition.

CameraPose LandmarksFeature / Motion RepresentationTensorFlow Lite Exercise ClassifierExercise + ConfidenceValidation GatesRep Counter
Personalized Learning Loop

The system did not only learn what a squat looks like. Over time, it learned what your squat looks like.

Instead of treating every user as if the system was seeing them for the first time, completed exercise sessions generated valid movement data for a separate training/adaptation pipeline. The app used TensorFlow Lite for inference; training and adaptation happened separately, and updated personalized models were used for future inference.

Range of motionSpeedBody proportionsJoint trajectoriesExercise style
Generic Model->User Performs Exercise->Valid Movement Data->User-Specific Training / Adaptation->Personalized Model->Improved Recognition
Final Exploit

Playing a Squat Video

Eventually, users tried playing prerecorded squat videos in front of the camera. The camera could still see a person, correct body landmarks, and valid squat motion. Since standard Android devices do not provide reliable universal depth sensing that simply solves this, I avoided expensive screen detection or monocular depth estimation and introduced a device-sensor / physical-context handshake using Android APIs.

When one sensor can be fooled, validate it with an independent signal.
Defense in Depth

Camera + Device State

The vision result was no longer trusted by itself. The system required agreement between the camera / pose signal and an independent device physical-state / sensor signal before recognition was accepted. This acted as a lightweight liveness check without exposing sensitive anti-cheat implementation details.

FINAL ARCHITECTURE

Computer vision surrounded by context.

CONCEPTUAL / NON-PROPRIETARY
UserCamera InputLighting Quality GateDevice Orientation / Sensor ValidationPose Landmark DetectionBody Posture ValidationTFLite Exercise ClassificationAnti-Cheat / Liveness ValidationRep DetectionExercise ResultUser-Specific Training DataPersonalization PipelineImproved User Model
Engineering Lessons

What This Project Taught Me

Production engineering is iterative. A feature can become a learning system instead of remaining static.

Start with the simplest solution that solves the problem.Let real-world data expose incorrect assumptions.Users interact with systems in ways developers never predict.Anti-cheat systems work better when independent signals are combined.Do not introduce ML simply because ML sounds impressive.Move from deterministic rules to learned models when complexity justifies it.Reject unreliable input early.Technical ownership means understanding why an architecture evolved.
Open Full Technical Breakdown

Problem

Detect camera-based exercise movement without claiming ownership of pose estimation itself. The algorithm consumed labeled skeletal landmarks and reasoned about exercise motion, validity, and rep count.

Initial Approach

Joint trajectories were tracked over time and compared against the expected squat cycle. The waveform confidence threshold gave the first working version a simple, inspectable decision rule.

Real-World Failures

Background session recordings revealed lying down, phone rotation, poor lighting, jittery landmarks, and prerecorded video attempts. Each one exposed a wrong assumption about the input stream that normal development testing had not produced.

Iterations

Each failure added the smallest useful validation layer: standing posture, skeletal landmark orientation checks, phone orientation, lighting quality, Android sensor/physical-context validation, and eventually learned movement classification.

Final Architecture

Recognition became a validation-gated system: camera input, environmental gates, existing pose landmarks, posture validation, TFLite inference, sensor-based liveness checks, rep counting, and a separate personalization/training pipeline.

Tech Stack

KotlinAndroid SDKCameraPose / skeletal landmark detectionTensorFlow LiteAndroid SensorsDevice Orientation APIsComputer VisionMachine LearningPersonalized model pipelineActivity / exercise recognition
04 / Case notes

Engineering details
worth inspecting.

Concise project notes with role, challenge, technologies, and links.

Open source / Gradle Plugin

ArchGuard

ArchGuard addresses architecture drift in growing Android projects by validating feature-first structure, Clean Architecture boundaries, and executable architecture rules.

It is Kotlin and Gradle developer tooling designed to make project constraints visible through repeatable checks and HTML reports.

KotlinGradle PluginFeature-first validationClean ArchitectureHTML reportsDeveloper tooling
Explore ArchGuard on GitHub
Open source / Android SDK

Biometric SDK

Biometric SDK is a reusable Android authentication SDK for projects that need lifecycle-aware BiometricPrompt integration behind a cleaner developer API.

The work reflects reusable SDK design, Android lifecycle awareness, and packaging implementation details into simple integration contracts.

KotlinAndroid SDKBiometricPromptLifecycle-awareReusable APIs
Explore Biometric SDK on GitHub
Android app

Sally Launcher

Sally Launcher is a custom Android launcher focused on a clean, smooth user experience and fast access to core phone actions.

It represents practical Android SDK work, Kotlin implementation, and product thinking around everyday mobile ergonomics.

Android SDKKotlinUXLauncher
View Sally Launcher on Google Play
05 / Flutter positioning

From Native Android
to Flutter.

Hands-on Flutter built on years of production Android architecture experience.

Architecture transfers across platforms

After years of production Android engineering, I expanded into Flutter by migrating Tapori AI to a cross-platform client while preserving the same Clean Architecture principles, backend contracts, and product behavior.

This is hands-on Flutter experience grounded in senior mobile engineering, not a claim of 6+ years of Flutter-only work.

Android ViewModel->Riverpod Notifier
Retrofit / OkHttp->Dio
Navigation Component->GoRouter
Repository->Repository
Use Cases->Use Cases
06 / AI engineering

Mobile Engineering
Meets AI.

AI features treated as product systems, not one-off demos.

Typical AI path

Mobile UI↓State holder↓Use case↓Repository↓Supabase Edge Function↓OpenAI / LLM↓Grounded response
07 / Experience

Senior mobile work,
measured by ownership.

Jul 2020 - May 2026
Senior Android Developer

Vantage Circle

Guwahati, India

  • Owned Android engineering across enterprise products including Vantage Circle and Vantage Fit.
  • Migrated more than 60% of a legacy MVP codebase to modern MVVM architecture.
  • Converted 25+ XML screens to Jetpack Compose and supported Material 3 adoption.
  • Helped shape feature-based modular architecture, reusable components, and maintainable Android architecture boundaries.
  • Worked on Android performance optimization, profiling, and modernization of older app layers.
  • Implemented motion and fitness-related engineering work involving Kalman filtering and ML Kit Pose Detection.
  • Migrated persistence from SQLite patterns toward Room where appropriate.
  • Supported Android 15 migration/readiness and platform compatibility work.
  • Mentored engineers, reviewed architecture decisions, and took technical ownership across product delivery.
KotlinJetpack ComposeMVVMMaterial 3RoomML KitPerformance
2018 - 2020
Mobile & Web Developer

GeekWorkX Technologies

Guwahati, India

  • Delivered multiple Android applications for clients using Java and Kotlin.
  • Built cross-platform applications using React Native.
  • Developed REST APIs and backend services using PHP and MySQL.
  • Worked across the full software development lifecycle, from requirements gathering to deployment and support.
JavaKotlinReact NativePHPMySQL
2018
Android Developer

KBG Software

Guwahati, India

  • Developed Android applications for educational institutions.
  • Implemented attendance management, notifications, student records, and communication features.
  • Collaborated directly with clients to gather requirements and deliver production-ready solutions.
08 / Skills

Grouped by
engineering use.

A practical stack across mobile, architecture, backend integration, AI, and tooling.

Mobile

AndroidFlutterKotlinDartJavaJetpack ComposeMaterial 3Android SDK

Architecture

Clean ArchitectureMVVMMVISOLIDFeature-first architectureModularizationMulti-module architectureRepository pattern

State & Async

RiverpodCoroutinesFlowStateFlowRxJavaDart async/await

Networking & Backend

DioRetrofitOkHttpRESTGraphQLSupabaseFirebaseKtorWebSockets

AI

OpenAILLM integrationAI-assisted applicationsSupabase Edge FunctionsPrompt-driven featuresRAG-style professional context

Tooling

GradleGitHub ActionsCI/CDAndroid ProfilerGitMCPDeveloper tooling
10 / FAQ

Frequently asked
questions.

Short answers for recruiters and engineers scanning the portfolio.

Who is Nizamuddin Ali Ahmed?

Nizamuddin Ali Ahmed is a Fullstack DevOps Engineer based in Guwahati, Assam, India, with 6+ years of mobile and Android engineering experience and hands-on Flutter work.

What does Nizamuddin specialize in?

He specializes in Android, Kotlin, Jetpack Compose, Flutter, Dart, Clean Architecture, Android architecture, Mobile AI, LLM integration, Supabase, SDKs, Gradle plugins, and developer tooling.

What is Digital Me?

Digital Me is an AI-powered interactive version of his professional profile. Visitors can ask about experience, projects, Android expertise, Flutter work, architecture decisions, and career background.

What has he built with Flutter?

He migrated Tapori AI to a Flutter client using Dart, Riverpod, Dio, GoRouter, and Clean Architecture while preserving backend contracts and product behavior.

What is Tapori AI?

Tapori AI is an Android and Flutter AI chat application with contextual chat, authentication, credit management, payments, Supabase Edge Functions, OpenAI integration, and LLM-backed flows.

What is ArchGuard?

ArchGuard is an open-source Gradle plugin for executable architecture rules, feature-first validation, Clean Architecture enforcement, and HTML reports.

What is SonicBridge?

SonicBridge is a TV-to-mobile low-latency audio streaming system using Android TV, AudioPlaybackCapture, WebSockets, PCM audio, AudioTrack, and local network streaming.

How did users cheat the squat detector?

Recorded testing sessions showed users lying down, changing phone orientation, and later trying prerecorded squat videos. Each attempt exposed an assumption, which led to posture validation, Android orientation checks, lighting gates, and sensor-based liveness validation.

Is he only an Android developer?

No. Android is his deepest production background, but the current portfolio positions him as a Fullstack DevOps Engineer working across Android, Flutter, architecture, AI integrations, and developer tooling.

11 / Contact

Need software
engineering depth?

I'm open to software engineering conversations across Android, React Native, web, cloud, AI integrations, developer tooling, and production product work.

07 / Education

A different starting point,
the same curiosity.

2013 — 2017

Bachelor of Engineering
(Mechanical Engineering)

Royal School of Engineering & Technology, Guwahati, Assam

2010 — 2012

Higher Secondary (Science)

Sankardev Academy, Nalbari, Assam · Completed with 78%.

2009

Secondary School (Matriculation)

Dreamland English School, Mukalmua, Assam · Completed with 81%.

Final year projects included Stress Analysis of a Bicycle Frame and Design of a Klann Walking Mechanism (Hexapod Robot).

08 / What matters

Good work is
shared work.

“Thoughtful engineering is the habit of making the next change easier than the last.”

— A principle I bring to every team

“The best mobile experiences disappear into the moment they make possible.”

— Product-minded Android engineering
09 / Device lab

Designed for
every surface.

Android experiences should feel native wherever they land.

09:41 100%
Sally LauncherCalm software for focused days.
DEVICE / PIXEL 9 PRO

Tap a device profile to inspect the responsive surface.

10 / Live architecture

Systems that
stay legible.

01PresentationCompose · UI state
02DomainUse cases · contracts
03DataRepository · mapping
04ANetworkGraphQL · Retrofit
04BDatabaseRoom · DataStore
NEXUS / ENGINEERING BEYOND SOFTWARE

Technology changes.
Engineering thinking remains.

I believe engineering exists to solve real-world problems. Sometimes that solution is software. Sometimes it is hardware, artificial intelligence, or system design.

01 / SOFTWARE ENGINEERING

Make the next change easier.

SYSTEMS / ONLINE
ENGINEERING PRINCIPLES
Presentation↓Domain↓Data↓API / Database
02 / SONICBRIDGE

Private TV Audio.
Reimagined.

ANDROID SYSTEM / RESEARCH
Engineering private entertainment through software.

An Android-powered low-latency audio streaming system that turns a TV into a personal wireless listening experience—over local Wi-Fi, without cloud services or expensive hardware.

Android TVCapture
→
Local Wi-FiNo cloud
→
Android PhonePrivate playback
THE ENGINEERING CHALLENGE

Late-night TV should not be a negotiation.

Bluetooth introduces latency. Wired headphones reduce comfort. Premium wireless systems are expensive. SonicBridge asks whether software can solve the real problem with devices people already own.

WHY LOCAL WI-FI?

More bandwidth, more control, and no internet dependency—enough room to move stereo PCM audio with a low-latency buffer.

LOW LATENCYSTEREO PCMOFFLINE OPERATION
ENGINEERING FOCUS / SYSTEM PATH

From capture to comfort.

LOCAL / DIRECT / PRIVATE
IMPLEMENTATION NOTES

An engineering case study, not an app description.

03 / AI RESEARCH LAB

Questions before products.

NEURAL FIELD / EXPLORING
REASONING
WITH CARE
PROMPTS
AGENTS
VISION
HUMAN + AI
03 / PRODUCT LAB

Ideas waiting for a first prototype.

CONCEPTS / DRAFT
04 / OPEN SOURCE

Tools that make teams faster.

Open GitHub mission control
NONCODERF
REPOSITORIESACTIVE
CONTRIBUTIONSOPEN
TOOLINGGRADLE / ANDROID
05 / ENGINEERING JOURNAL

Write down the hard-won parts.

Read the research
FIELD NOTES / MEDIUM
NIZAMUDDIN ALI AHMED / ENGINEERING NOTES

Architecture is a communication problem.

Technical writing is where implementation details become shared context. Explore notes on Compose, performance, clean architecture, and building systems that teams can understand.

Open knowledge base
06 / EXPERIMENTS

Curiosity is a technical skill.

NOTEBOOK / UNFILTERED
01 / FAILED IDEA

Too much abstraction

When a system hides every decision, it also hides the path to a better decision.

Lesson / Make boundaries visible.
02 / PROTOTYPE

Ambient interfaces

Exploring interfaces that communicate status through light, motion, and restraint.

Lesson / Feedback can be quiet.
03 / RESEARCH

Physical + digital

Software thinking can shape a physical experience.

Lesson / Systems have edges.
07 / LIVING ENGINEERING TIMELINE

One problem leads to another.

08 / ROADMAP

The work keeps opening up.

NEXUS / ENGINEERING COMMAND CENTER

I didn't just meet an Android developer.
I met an engineer who solves problems.

15 / Connected journey

From Recipes
to Repetitions.

Two everyday problems, connected by the same instinct: build what you wish existed.

FindMyRecipe solved one everyday problem: what should I cook today?

Then came a full stomach, a comfortable bed, and absolutely no motivation to exercise. After seeing “Unlock Premium” and “Subscribe to Continue” everywhere, the same thought returned: I'm an Android developer. Why not build one myself?

CHAPTER 01

FindMyRecipe

AI-powered ingredient recognition and recipe discovery, built around a question from my mom.

Explore the project
CHAPTER 02

FormFit

Movement tracking, pose analysis, repetition logic, and a vision for Record & Repeat.

Read the case study
16 / FormFit

Android / Computer Vision / Fitness Technology

Move. Measure.
Improve.

FormFit is an AI-powered fitness and human-movement tracking app built to make exercise feedback useful without putting the essential experience behind a subscription.

KotlinJetpack ComposeCameraXML Kit Pose DetectionTensorFlow Lite
FormFit artwork showing pose landmarks across exercises and Record and Repeat
THE ORIGIN STORY

First I built an app to decide what to eat. Then I needed one to burn it off.

I wanted workout tracking without essential features hidden behind premium subscriptions. I also love dancing—not professionally, but as a bedroom dancer copying videos on the television. Remembering every movement and getting the timing right is hard.

That led to a bigger idea: Record. Repeat. Improve. Record & Repeat is presented as an evolving product direction, not as a claim that every movement is already shipped.

FEATURES / SHIPPED + EXPLORATORY

Movement feedback
in context.

Screenshots show the product surface; labels distinguish available flows from the broader direction.

FormFit push-up tracking screen
AVAILABLE FLOW

Push-up tracking

Pose-based movement tracking with live repetition state and session controls.

FormFit squat and push-up tracking screens with pose landmarks
AVAILABLE FLOW

Squat tracking

Body landmarks support movement analysis, repetition counting, and visible form feedback.

FormFit workout selector for push-ups, squats, jumping jacks, and treadmill running
AVAILABLE SURFACE

Jumping jacks and treadmill running

The selector presents movement-specific tracking surfaces and available metrics without overstating internal implementation.

FormFit custom movement Record and Repeat screen
EXPLORATORY DIRECTION

Record & Repeat

Capture a reference movement and practice it again through pose analysis—the idea connecting fitness tracking with dancing.

TECHNOLOGY STACK

Android UI, vision, and movement logic.

ANDROID UI & ARCHITECTURE

Kotlin · Jetpack Compose · Android SDK · MVVM · Clean Architecture

COMPUTER VISION

CameraX · frame analysis · ML Kit Pose Detection · body landmarks

MACHINE LEARNING

TensorFlow Lite · on-device inference surface · confidence-aware analysis

EXERCISE LOGIC

Movement detection · landmark processing · repetition counting · feedback state

17 / FormFit architecture

From camera frames
to useful feedback.

A lightweight pipeline: capture, detect, analyze, count, and surface state in Compose.

01Camera / reference videoCameraX capture or movement input
02Frame acquisitionLifecycle-aware image analysis
03Pose detectionML Kit landmarks per frame
04Movement analysisLandmark features and exercise logic
05Rep count / feedbackConfidence and UI state
BEYOND THE UI / ENGINEERING FORM FIT

Problem → Investigation → Solution → Learning

Real-time processing

Problem: camera analysis must not block the UI. Investigation: frame acquisition and pose work need a lifecycle-aware analysis path. Solution: keep processing separate from Compose state and surface meaningful results. Learning: responsive feedback depends on disciplined boundaries.

Pose reliability

Problem: landmarks vary with framing, lighting, and confidence. Investigation: movement logic needs more than one raw coordinate. Solution: process landmarks into movement signals and treat confidence as input. Learning: reject weak signals before counting.

Release-build dependency investigation

Problem: an older WorkManager multiprocess dependency appeared in the release runtime classpath investigation. Investigation: dependency resolution and runtime packaging must be checked together. Solution: this portfolio records the investigation without claiming a verified root-cause fix that is not visible in the supplied frontend/repository context. Learning: release artifacts are part of feature ownership.