I build software because I like to see what consumer hardware can actually do when you refuse to accept predefined limits. Everything here was engineered to solve my own bottlenecks and test the boundaries of on-device AI, physical sensors, and edge runtimes. One of them, Forensic Insight Hub, is already live and free to use in the browser. I'm still polishing the rest of the arsenal, and when each one is ready, it will be released completely free without subscriptions, paywalls, or tracking.
I don't start with institutional assumptions about what a phone or laptop can do. If an idea seems too heavy for consumer hardware, I try to run it locally: the GPU for language models, the NPU where an embedding model actually maps onto it, and SIMD for tight numeric loops.
Your data belongs on your glass. The phone tools are built on-device and offline-first, or route directly to your chosen AI provider using your own key. Forensic Insight Hub is the candid web exception: text goes to OpenRouter with the lab’s limited shared key through our edge function, or with your own key from the browser. No tracking, no ads, and no selling your data.
I build tools to sharpen daily life and deliberate practice, not to manufacture dopamine addiction or monthly subscription fees. One is already free to try in the browser; the rest will be published freely once each is thoroughly tested.
Here are some of the things I've built. Each project is explained in plain English alongside its underlying technical architecture. Click any card to explore the full story, screenshots, and build notes.
Try in browser: Forensic Insight Hub ↗ · read the story
Paste a privacy notice, contract, or terms of service and get it back in plain English: who can receive your data, where the power sits, what each clause really means, and what to push back on. Every flagged quote is checked against the original text. It is public at insight.necessitylabs.com.
“We don’t promise absolute data privacy. Isolation is hard, and a prompt can go straight to a provider with your own key. What stays on the phone is a SQLCipher database whose key is sealed in the Android Keystore. There is no server of mine in between, and the apps don’t ship tracking SDKs. Local language models run through LiteRT on the GPU. The embedding model is the one that uses the NPU.”
“Most software is constrained by institutional assumptions about what a phone or edge runtime can do. Because Necessity Labs isn’t beholden to conventional roadmaps or commercial playbooks, we don’t start with preconceived limits. If a workflow or local architecture seems impossible on mobile hardware, we don’t dismiss it. We find the mechanical workaround to make it run. When you don’t accept predefined boundaries, the surface area of what you can build expands dramatically.”
Native Android runtimes, a hybrid local-cloud simulator, Forensic Insight Hub on the public edge, and Necessity Hub, the private edge command center.
| Domain | Runtime / Stack | Implementation Focus |
|---|---|---|
|
Edge Intelligence
|
LiteRT-LM, Microsoft ONNX Runtime Android, LiteRT GPU Delegate, MobileCLIP, Whisper
|
LiteRT language models run mainly on the GPU. The embedding model is the path that uses the Qualcomm Hexagon NPU. ONNX Runtime Android runs the MarianMT translation model. Recurrent anomaly models, including GFP-LSTM, use the LiteRT GPU delegate.
|
|
Native Android
|
Kotlin, Jetpack Compose, Coroutines, Room, SQLCipher, C++/JNI
|
Monastic OLED UI, custom canvas graphics, Keystore key sealing, hardware scan code interception, background monitoring services.
|
|
Sensor Fusion & Native Systems
|
C++ / JNI, ARM NEON SIMD, Android Sensors API, CMake / NDK
|
Android delivers accelerometer, gyroscope, barometer, and light events asynchronously. The app time-aligns them. SIMD covers the tight numeric path. The particle field is drawn on a Compose Canvas, on the UI thread.
|
|
Hybrid AI & Simulation
|
Godot 4.7, Python 3.12, bashkit, LiteLLM / OpenRouter, Local SLM Runtimes
|
bashkit's virtual filesystem stays off the host disk. Model calls are separate: a local Gemma server for ordinary turns, then DeepSeek V4.1 Flash through OpenRouter when a run escalates. The observer draws the run as a topology graph.
|
|
Forensic Insight Hub
|
Vite, React, TypeScript on a Cloudflare Worker, Supabase Auth / Postgres / Edge Functions, OpenRouter
|
Public legal-text analyzer at insight.necessitylabs.com. A Supabase Edge Function streams the analysis from a curated model list or the user's own key, enforces a daily shared-key limit, and every quoted finding is matched back to the source before it is marked verified.
|
|
Necessity Hub
|
Workers, Hono, React 19, Agents SDK, Durable Objects, Workflows, D1, KV, R2, Queues, Vectorize, AI Gateway
|
Private command center at hub.necessitylabs.com. Stateful agents, a nine-phase pipeline with approval gates, outbound device tunnels, and an MCP server for local coding tools.
|
|
Data Architecture
|
Local SQLite, JSON Schemas, Markdown Vaults
|
Portable document templates, bidirectional Obsidian wikilinks, local-first storage without vendor lock-in.
|
Extensive floor-level manufacturing, logistics, customer service, and operational lead roles. Deep practical experience solving physical bottlenecks where theory meets reality.
Learned by building and shipping production software: native Android/Kotlin, on-device AI/ML acceleration, and distributed Cloudflare edge systems. Currently working in an entry-level development role while independently engineering the Necessity Labs arsenal.
Twenty years of building things and trying to make a living from it. That's the actual credential.
Not credentialed in every field, but proven repeatedly to get functional in unfamiliar domains fast. The drive doesn't turn off when the subject changes.
Nothing here is for sale. One tool is already public and free to try in the browser: Forensic Insight Hub. The Android apps are still being packaged for free release (no ads, no paywall). If any of this is useful, voluntary contributions help pay for hardware experiments.
Necessity Labs is driven by continuous learning and pragmatic execution. I build tools that challenge the line between consumer devices and autonomous edge infrastructure. If you have a challenging technical problem, an unconventional project, or an opportunity that demands relentless problem-solving, reach out:
Opens your email app, addressed to contact@necessitylabs.com.