GLYPH.

The problem

Real life breaks meal plans daily. Office pizza appears. Social drinks happen. Workouts shift. 87.5% of users face 2–8+ unplanned food events per month — and every existing nutrition app's response is the same: log it, see the damage, feel bad, try again tomorrow.

The deeper failure is structural: nutrition apps are systems of record — they capture what you ate, after the fact, and leave you to figure out what to do next. The math required to actually rebalance the remainder of the day (protein remaining, fat ceiling, carb window around training) is exactly the math nobody does when they're tired, stressed, or already feeling like they failed.

The solution

GLYPH inverts the model from record to action. The user snaps a photo of what they actually ate; the Multimodal Intake agent identifies the food and macros; the Agentic Rerouter recalibrates Meals 3 and 4 in real time to stay within daily targets — without dropping protein floors or training-window carbs. No log-and-shame. No "you went over." Just a rerouted day, GPS-style.

The system explicitly frames every deviation as new input, not failure — collapsing the emotional cost of micro-failures and keeping the metabolic habit loop intact through the events that normally kill it.

How it works

The Navigator Loop: Morning Navigator sets daily context → user executes plan → deviation ("sigh moment") occurs → AI Vision identifies the deviation → AI Rerouter recalibrates the remaining day autonomously → notification + user accepts/overrides.

Orchestrator-worker multi-model topology: GPT-4o for Planning (complex JSON adherence, deterministic structure); Gemini 2.0 Flash for Vision (food image → structured macros); with deterministic fallback to a secondary provider per sub-agent. Each model is matched to its core competency, not chosen as a single default.

Prompt architecture: layered S.P.C.R. (System, Persona, Context, RAG, User, Schema) with negative constraints first, XML-tagged blocks, and a self-verification footer. Persona: The Clinical Navigator — Sports-Med PA voice, calm, precise, 2nd person, no emojis. Safety reasoning also runs across diet/supplement conflicts (e.g. Zinc ↔ Oyster interactions).

Who it's for

Primary target audience: "Busy Tech Professional" — high-pressure schedule, frequent unplanned food events, low tolerance for manual logging, comfortable with AI tools, willing to pay for cognitive offload.

Niche secondary: GLP-1 medication users (Ozempic, Wegovy, Mounjaro) — a fast-growing demographic with real-time nutritional management needs their prescribing physician can't service on demand. The agentic rerouting model fits GLP-1 use cases especially well, since appetite changes are unpredictable and protein floors are clinically important.

Why it matters

Personalized nutrition is on track to $35.96B by 2031 at 14.06% CAGR — and two waves are colliding: the GLP-1 medication boom is creating millions of new users who need precise, real-time nutritional management, and agentic AI maturity has finally crossed the threshold where autonomous re-planning is reliable enough for daily use.

GLYPH's structural bet is that "System of Action, not Record" is the right framing for the entire next generation of health apps — the user doesn't want to log their life, they want their life logged for them and the math done in the background. Whoever wins the agentic rerouting primitive wins the category.

At a glance

Project

GLYPH

Built by

Gregy Thomas Kokkaparampil

One-liner

Agentic metabolic nutrition navigator with a Snap→Identify→Recalculate→Notify rerouting loop — GPS for your macros, not a diet that judges.