RH

things i've shipped

solo builds. real deploys. some have users; all have git history.

research snapshot (no PDF — just the story)

Problem. Roughly 57 million people live with dementia; many wait years for an accurate diagnosis. Frontotemporal dementia is misdiagnosed as a psychiatric disorder in up to half of cases — wrong label, wrong treatment, years lost.

Approach. C‑MAT: a cross-modal attention system that jointly uses brain imaging and EEG-style signals to distinguish Alzheimer’s disease, frontotemporal dementia, and Parkinson’s — designed to stay useful when part of the data is missing.

Why it matters. Training landed around ~$6 on commodity setups; evaluated across 1,102 subjects, 10 datasets, and 7 countries. Not a replacement for neurologists — a faster second opinion. Still early research, but directionally promising.

C-MAT architecture: MRI + EEG → Cross-Modal Attention → 4-class classification
C-MAT pipeline — structural MRI + resting-state EEG → cross-modal attention → AD / FTD / PD / Healthy

Products

ChannelSpy preview

ChannelSpy

product

YouTube competitor intelligence — for people who hate guessing

Converts any YouTube channel URL into a full performance report — KPIs, trend charts, estimated earnings, and CSV export in ~15 seconds. Server-side route handlers protect API keys. Median-based scoring with IQR outlier detection means one viral video doesn't skew the whole benchmark. 11-event GA4 funnel — because "users visited" isn't analytics.

Next.js 15 · TypeScript · YouTube Data API v3 · Recharts · GA4 · Vercel

11-event GA4 funnel · IQR outlier scoring · ~15s to full report

GitHubLivecase study →
Ummah Speaks preview

Ummah Speaks

product

AI Islamic companion — emotional intent, sub-2s latency

Detects what you're actually feeling (not just what you typed) and responds with personalised Islamic reflections via Llama 3. In-memory rate limiting so a weekend of traffic doesn't cost $400. Privacy-first local storage — some conversations aren't for the cloud, and that's not a bug.

Next.js 15 · TypeScript · Groq API · Llama 3 · Tailwind CSS 4

sub-2s latency · emotional intent · zero server-side storage

GitHubLivecase study →
SkipTheTerms preview

SkipTheTerms

product

AI Terms of Service summarizer — because life is short

Chrome Extension + FastAPI backend that reads the Terms of Service so you don't have to. Llama 3.3 70B in under 3 seconds cold. Supabase caching on document hash: cache hits return in 50ms, cutting LLM costs 40%. The best LLM optimization is not calling the LLM.

FastAPI · Python · Manifest V3 · Supabase · Groq API · Llama 3.3 70B

50ms cache hits · 40% cost reduction · <3s cold

GitHubLive: Coming sooncase study →