The fastest AI value comes from inserting capability into workflows people already inhabit — not from asking them to learn new behavior for a paradigm shift that hasn't earned trust yet. Meet users where their hands already are.
Grammar correction, autocomplete, and summarization are commodities. Durable advantage lives one layer below: deeper context capture, proprietary behavioral data, or a system-layer position competitors can't reach without rebuilding the stack.
Skip it once and you pay it back later, with interest. Build a product on top of an audience that doesn't exist on the platform you chose, and no amount of execution will rescue the thesis. Demand decides every technical direction of this era — and an AI product only lands where someone is actually willing to pay.
When something doesn't work, name the failure modes specifically — pricing model, beachhead miss, organizational risk, mental-model mismatch. Vague regret teaches no one. The next venture is funded by the lessons of the last.
Research-project mode and paid-obligation mode produce different companies. The job of early capital isn't growth — it's converting talented friends into a team with skin in the same game, on the same clock.
Founders fail by inertia more often than by bad luck. Decide in advance — in writing — what condition flips Plan A to Plan B to Plan C, and on what date. Future-you, mid-crisis, will not be the one to make that call cleanly.
A Delaware C-Corp, three engineering co-founders, and a single conviction: the input layer is the most under-built surface in AI.
The thesis is structural. Every assistant lives downstream of an intent the user has already half-formed; almost no product is paying attention to that moment. The leverage is upstream.
The current build is a Chrome extension — a deliberate choice over a native input method, because the semantic context available in the DOM is richer than what an OS-level IME can see. The cost is platform constraint; the trade is depth.
Operating in research-project mode for now, with a near-term friends round scoped to convert that into a real operating cadence. Knowing the difference matters.
Iterant AI, Inc. — Delaware
Incorporated the company, recruited the engineering team, set product architecture, and own the intent-classification system design. The work is moving from research mode to operating mode.
KeploreAI
Brought high-value connections from the Chinese automotive sector — Geely VP, Lotus CEO, a Yaoguang Tech case study — and recommended verticalizing around that beachhead. The recommendation wasn't taken.
Lesson taken forward: a SaaS pricing model layered onto deep-tech B2B can produce negative unit economics for a year before the math is visible. Pricing is strategy, not a knob.
University of Pennsylvania
Coursework across databases, AI, and big-data analytics — including CIS 5500, CIS 521, CIS 5450. STEM-designated; the formal training I needed to stop being a "non-technical founder" in any conversation that mattered.
Independent project
Built the operating thesis for an AI-driven talent-discovery model across CS2 / Valorant / MOBA, including investor mapping (BITKRAFT, Griffin, Makers Fund, a16z Games) and a tournament-driven brand strategy.