Not polished products — raw signals. These are ideas I'm turning over, problems I keep returning to, and observations that won't leave me alone.
What if the user always knew exactly what the AI could see?
Most AI interfaces obscure context — what's in the prompt, what's remembered, what's inferred. I want to explore a visible "permission card" UI layer that surfaces this in real time, giving users genuine legibility over what they're interacting with.
What would make this real? A browser extension prototype for any chat UI, surfacing token context and system-prompt scope as a readable sidebar.
Removing all friction isn't always the right move.
Design dogma says eliminate friction. But some friction builds meaning — the delay before a letter arrives, the effort of a handwritten note. I'm collecting examples where designed friction improves the experience, not degrades it.
What would make this real? A written essay + a small annotated collection of friction-forward products that feel better for being slower.
The same content shouldn't look the same at 7am and 11pm.
Reading mode tools strip pages bare but ignore context entirely. What if reading mode was ambient-aware — typography density, contrast, and pacing adjusted to estimated cognitive state, time of day, and session length?
What would make this real? A Firefox extension with time-of-day reading profiles and a small study measuring retention across conditions.
What if your portfolio was one perfect page, not thirty average ones?
Generalist portfolios reward breadth. But specialists — a single-domain engineer, a focused writer, a niche researcher — need a different format. Depth over breadth. One case study that says everything, rather than a gallery that says nothing.
What would make this real? A template kit and a written framework for choosing what single project best represents you.
Most apps pretend the internet always works. It doesn't.
Offline-first architecture is well understood technically. What's underexplored is the design language around it — how do you communicate sync state, conflict resolution, and uncertainty to a non-technical user without breaking their trust?
What would make this real? A design pattern library specifically for sync-state UI — status indicators, conflict modals, and latency-honest copy.
Could a subtle audio layer make interfaces feel more alive?
Explored using non-intrusive spatial audio cues — low-frequency tones, subtle clicks — to reinforce UI hierarchy and navigation depth. Dropped after testing: the overhead of audio permissions, accessibility conflict, and user habituation outweighed the benefit.
Why dropped? Accessibility requirements make opt-out audio patterns risky. The signal-to-noise problem wasn't solvable at low implementation cost.
What if logs told you what they meant, not just what happened?
Distributed systems produce mountains of logs that require expert knowledge to interpret. I'm prototyping a debug view layer that maps telemetry events to plain-language system state explanations — bridging the gap between raw observability data and actionable understanding, even for engineers triageing on-call at 2 AM.
What would make this real? A lightweight SDK that wraps existing log emitters and injects semantic annotation tags that a local rule engine renders into natural-language state summaries.
Could your ambient environment tell you something is wrong before you notice?
Most systems notify through screens. But screens demand focused attention. I'm thinking about low-bandwidth physical outputs — ambient LED gradients, micro-haptic pulses, subtle peripheral luminescence — that surface telemetry health passively without active monitoring overhead.
What would make this real? An open-hardware ESP32 desktop status beacon tethered to GitHub Actions / Prometheus webhook metrics.
What if every interface felt instant, regardless of actual network latency?
Perceived latency is not actual latency. Optimistic UI, predictive hover prefetching, and kinetic motion design can make a 300ms server roundtrip feel under 50ms. I'm cataloguing the full taxonomy — speculative execution, tactile press physics, and gesture-coupled animation — that eliminate felt waiting time.
What would make this real? A framework-agnostic micro-library for touch/hover intention prediction and optimistic state rollback.
Can procedural randomness feel deliberate and alive?
Generative motion often sits at two extremes: either too rigid and algorithmic, or too noisy and chaotic. I want to explore interface motion governed by deterministic pseudorandom seeds and physical constraints — motion that feels organic like handwriting, yet repeatable across browser sessions.
What would make this real? A seeded PRNG animation driver that syncs generative SVG and Canvas physics to high-frequency touch interactions.
This person doesn't wait for ideas — they generate them.