Product · Security · Featured
Zerova / Zystra VPN
StackFlutter · Dart · Kotlin · WireGuard
PlatformsAndroid (Google Play)
TypeVPN client · rebrand lifecycle
RoleMobile engineering · publishing
A cross-platform VPN client built on a WireGuard core, with native tunnel management, multi-region server selection, and a complete privacy and rebrand lifecycle — designed, built, signed and shipped to Google Play.
The problem
The client needed a real VPN — not a wrapper around a third-party SDK, but a proper tunnel with native service integration, reliable connection state and store-compliant privacy handling. It also had to survive a full rebrand from Zerova to Zystra without breaking existing installs.
What we built
- Native Android tunnel via VpnService with the WireGuard core, Flutter UI as a thin state reflector.
- Full connection lifecycle — connect, disconnect, reconnect-on-network-change and a foreground notification.
- Multi-region server selection with latency-aware defaults.
- VPN service declaration, data-safety form and privacy policy written to match real behaviour.
- Signing-key continuity preserved through the rebrand so Play treated the new identity as an update.
Outcome
A production VPN client delivered through the full store lifecycle, including VPN declaration, data-safety compliance and a rebrand handled without losing update continuity.
Automation · Finance
AI Trading Bot
StackPython · MQL5 · Pine Script · CDP
PlatformsMetaTrader · Web dashboards
TypeAutonomous research system
RoleSystems & ML engineering
A self-improving trading research system that records market data, learns a direction model, validates it honestly, and only trades horizons that pass a statistical edge test — with a live console read-out and layered risk controls.
The problem
Manual trading is inconsistent and emotional. The goal was a system that executes a defined strategy identically every time, tracks performance in real time, and limits downside automatically.
What we built
- Separated strategy (Python models) from execution (MQL5) so signal logic could evolve independently.
- A purged walk-forward validation gate that decides which horizons are tradeable.
- Risk controls: position sizing, maximum drawdown, daily loss caps and hard stop-losses.
- A live dashboard for PnL, open positions and strategy health, plus Pine Script indicators for chart validation.
- Full decision logging so strategy changes can be evaluated against real outcomes.
Outcome
A disciplined execution engine that runs the strategy consistently, with the risk layer enforcing limits no matter what the market does. Built and validated on a demo account.
Utility · Open Source
NetCarve — Subnet & CIDR Calculator
StackKotlin · Android
PlatformsAndroid
TypeOffline networking utility
RoleDesign · engineering · release
A free, ad-free, fully offline networking utility for network engineers, cloud architects and DevOps teams. NetCarve performs precise IPv4/IPv6 subnet math and VLSM planning entirely on-device using integer binary arithmetic.
The problem
Network engineers often work in air-gapped or restricted data centres where network access is unavailable — and most subnet calculators phone home, carry ads or track users.
What we built
- Exact subnet math derived from integer bit operations — correct for every prefix length.
- Requests no internet permission, contains no ads and no trackers.
- VLSM planning and a binary bit view with copy-to-clipboard on every value.
- Engineer-first UI with instant results, shipping under the MIT license as an open-source showcase.
Outcome
A genuinely private, offline-by-design tool trusted by engineers — and a public showcase of production-grade native software.
Reference · Mobile
AniEye — Offline Watch-Order Guide
StackKotlin · Android
PlatformsAndroid (min SDK 24)
TypeSingle-purpose reference app
RoleProduct · engineering · release
A fully offline anime watch-order guide: pick a series and see its recommended viewing order — seasons, arcs, films and optional entries, numbered step by step.
The problem
Fans constantly ask "which order should I watch this in?" but existing answers are buried in forums and wiki pages that require connectivity and are cluttered with ads.
What we built
- 14 popular series with step-by-step watch orders and 89 numbered steps.
- Optional steps (films/OVAs/spinoffs) clearly flagged, plus instant search by name.
- Zero permissions, no internet permission, no ads, no analytics, no stored data.
Outcome
A fast, private, ad-free reference app that works entirely offline — a clean example of shipping a focused product well.
AI · Mobile
AI-Powered Mobile Apps
StackOpenAI · Gemini · Flutter · Kotlin
PlatformsAndroid & iOS
TypeAI product development
RoleAI + mobile engineering
Android and cross-platform apps with integrated LLM features — conversational chatbots, image understanding, voice interaction and smart recommendations — built for real users rather than demos.
The problem
Clients wanted genuine AI capabilities without the fragility that comes from calling a model directly from UI code. AI calls fail, latency varies, and costs need watching.
What we built
- All model access wrapped behind a single AI service seam that owns prompts, retries, caching and fallbacks.
- Streaming responses so users see progress immediately.
- Graceful non-AI fallbacks so apps stay usable when the model is unavailable.
- On-device classification for cheap tasks, cloud models only where needed — cutting cost and latency.
- Instrumentation of the AI path: success rate, latency and fallback frequency.
Outcome
AI features that behave predictably in production — responsive under load, resilient to outages and cost-controlled through hybrid routing.
Full-Stack · Web
Full-Stack Web Platforms
StackReact · Next.js · NestJS · TypeScript
DataPostgres · MongoDB · Redis
TypeWeb products
RoleFull-stack engineering
Full-stack products with React/Next.js front-ends, NestJS/Node and TypeScript backends, and live deployment — built for real users with auth, data and payments.
The problem
Clients needed more than a landing page: real applications with authenticated users, persistent data, integrations and the reliability of a production deployment.
What we built
- Next.js/React front-ends with responsive, accessible UI and sensible SEO defaults.
- NestJS/Express and TypeScript APIs with typed contracts shared across the stack.
- Postgres/MongoDB persistence with migrations and Redis for caching and queues.
- Auth, role-based access and payment integrations where required.
- Containerised deployment with CI/CD and environment-based configuration.
Outcome
Deployed platforms that handle real traffic, with clean separation between front-end, API and data layers so features ship without regressions.
R&D · Autonomous
Defense Research & Development
FocusSimulation · signal processing · autonomy
DomainDefense R&D
TypeApplied research
RoleEngineering contributor
Our team contributes engineering to a national defense research and development project spanning simulation, signal processing and autonomous systems. Due to the sensitive nature of the work, specific implementation details are not published.
The problem
Complex defense research applications require rigorous engineering discipline, deterministic low-latency control logic, and high-reliability data parsing under harsh environmental uncertainties.
What we built
- Simulation of dynamic systems and evaluation of control strategies.
- Signal processing and radio tooling, including SDR-based experimentation.
- Autonomous-system behaviours — sensing, path planning and control.
- Applied research support and technical documentation.
Outcome
This R&D sharpens the same skills we bring to client products: rigorous simulation before shipping, signal and data processing, and autonomy that behaves predictably under uncertainty.
Embedded · Systems
Embedded & Firmware
StackESP32 · PlatformIO · Embedded C/C++
FocusSensors · telemetry · radio
TypeEmbedded systems
RoleFirmware engineering
ESP32 and microcontroller firmware, sensor and telemetry integration, SDR and radio tooling — applied across client prototypes and research experiments.
The problem
Connecting hardware prototypes to cloud services often suffers from unstable firmware loops, power-drain bugs, and unhandled wireless reconnection states.
What we built
- ESP32/Arduino firmware with PlatformIO toolchains.
- Sensor drivers, telemetry pipelines and hardware integration.
- SDR and radio signal tooling for experimentation.
- Bridging embedded devices to cloud and mobile apps over Wi-Fi/BLE.
Outcome
Reliable device firmware that connects the physical world to the software products we build.
Healthcare AI · Product Engineering
Healthcare Document AI
StackPython · FastAPI · Next.js · Vector search
FocusRAG · document workflows · privacy
TypeAI product surface
RoleProduct & AI engineering
A privacy-aware AI product surface that turns unstructured healthcare documents into structured, reviewable data — built around retrieval, evaluation and deployment constraints rather than a demo chatbot.
The problem
Healthcare document workflows were slow and error-prone: staff read long unstructured records by hand, and any AI assistance had to operate inside strict privacy and deployment constraints without inventing facts.
What we built
- A retrieval pipeline over unstructured medical documents with explicit citation surfaces.
- Evaluation harnesses so answer quality could be measured instead of assumed.
- A review-first interface: AI drafts, humans confirm, every step is traceable.
- Privacy-aware backend architecture and deployment boundaries agreed up front.
- Incremental rollout with logging to compare AI-assisted output against the manual baseline.
Outcome
An AI-assisted workflow that keeps humans in control, makes uncertainty visible, and can be defended in a regulated environment — the same discipline we apply to every AI feature we ship.
Automation · Internal Tools
Operations Automation Platform
StackNext.js · Node.js · Python · Agents
FocusInternal tooling · agentic workflows
TypeOperations platform
RoleFull-stack & automation
Repeated operational work converted into structured internal tools and AI-assisted workflows — the unglamorous software that removes hours of manual effort every week.
The problem
Teams were losing hours to copy-paste work across spreadsheets, inboxes and third-party dashboards. The work was repetitive, easy to get wrong, and invisible until something broke.
What we built
- Structured internal tools that replaced the spreadsheet-and-inbox workflow.
- AI-assisted steps for extraction, classification and drafting, with human approval gates.
- Automations with retries, logs and explicit failure states instead of silent breakage.
- Role-based access so the right people see the right operational data.
- A dashboard showing throughput and where time was actually being spent.
Outcome
Operational busywork turned into a system with owners, logs and measurable throughput — freeing the team to spend time on work that needed judgement.
E-commerce · Product Design
E-commerce Storefront
StackNext.js · React · Responsive UI
FocusProduct discovery · conversion UX
TypePublic storefront
RoleProduct engineering & design
A polished public storefront with product discovery, brand presentation and purchase-oriented UX — designed and built end to end for a direct-to-consumer brand.
The problem
The brand needed a storefront that felt as considered as its products: fast on mobile, clear about what is being sold, and structured so a buyer can move from interest to checkout without friction.
What we built
- A responsive storefront with product discovery and strong brand presentation.
- Purchase-oriented layouts: clear pricing, hierarchy and calls to action.
- Performance work targeting fast mobile loads and stable layout.
- SEO-ready metadata, structured content and shareable previews.
- A design system the brand could extend without redesigning every page.
Outcome
A storefront that looks the part and converts — shipped as a public build with the design system documented for future pages.
Productivity · SaaS
Productivity & Task Product
StackNext.js · TypeScript · Node.js · MongoDB
FocusTask workflows · SaaS billing
TypePublic product build
RoleFull-stack product engineering
A focused productivity product built around goals, tasks and progress clarity — including SaaS foundations: accounts, Stripe payments, data models and deployment.
The problem
Task tools tend to accumulate features until they become another chore. The goal was a small, opinionated product where progress is obvious and the interface stays out of the user's way.
What we built
- Goal → task → progress flows with a deliberately small feature surface.
- Secure authentication and per-user data boundaries.
- Stripe payments and subscription state handled server-side.
- A modular monolith architecture that keeps the codebase understandable.
- Cloud deployment with monitoring and a straightforward release path.
Outcome
A shipped product with real accounts, real billing and a codebase a small team can keep evolving without a rewrite.
SaaS · Backend Architecture
SaaS Platform Foundation
StackTypeScript · Node.js · Next.js · Stripe
FocusBackend boundaries · product workflows
TypeSaaS platform
RoleFull-stack engineering
Production SaaS foundations with clear backend boundaries — auth, data modelling, payments and product workflows designed so features can ship without destabilising the platform.
The problem
Early SaaS codebases often collapse under their own shortcuts. This engagement needed foundations that could carry a real product roadmap: predictable data, clear module boundaries and payments that behave.
What we built
- A modular monolith with explicit module boundaries instead of tangled endpoints.
- Typed contracts shared between front-end and backend.
- Authentication, session handling and permission checks at the boundary.
- Stripe integration with webhook handling and idempotent subscription logic.
- Cloud infrastructure and CI/CD sized for a small team.
Outcome
A platform foundation where new features slot into place, and where the safest place to add logic is also the obvious one.
Mobile UX · Design Systems
AI Writing App — Design System
StackFigma · Design systems · Prototyping
FocusMobile UX · AI interaction design
TypeDesign archive
RoleProduct & UI design
Mobile-first product design for an AI writing assistant: keyboard interactions, flow design and a visual system explored end to end before a line of production code.
The problem
AI writing tools live or die on interaction detail — how suggestions appear, how the keyboard behaves, how much the interface interrupts the act of writing.
What we built
- Mobile-first flows for assisted writing, revision and suggestion review.
- Keyboard-aware interaction patterns prototyped in Figma.
- A visual system with reusable components and type/colour scales.
- Interaction states for empty, loading, error and long-content cases.
- Handoff-ready specs so engineering and design share the same language.
Outcome
A design archive that shows how we approach AI product design: prototype the interaction first, then build the smallest system that supports it.
