Prototype · 7 agents · simulated
Walk the applicant's journey, one agent at a time.
Admissions OS orchestrates seven AI agents around a single shared object — the Narrative Anchor. This prototype lets you step through the end-to-end workflow an MBA applicant would experience, from first intake question to final school comparison. The NAO builds as you progress.
1
One shared object
The NAO travels with the applicant across every agent. Claims made on page 2 are cited on page 7.
7
Specialized agents
Each does one job well. Foreground agents run on Haiku; the rest use Sonnet.
~$17
Per-student cost
Blended AI cost with prompt caching on, based on the economics section of the project plan.
The seven steps
-
Intake 30 questions · form onlyCapture the applicant's raw material — background, goals, constraints, voice.~0 cost
-
Strategy Sonnet 4.6 · NAO generatorRead the intake, produce the v1 Narrative Anchor — themes, claims, fingerprint.~$0.05 · 60s
-
Essay editor Haiku · foreground pipelineStory Architect checks C-C-A-I live. Editing agent flags weak verbs and fillers.~$0.002 / call
-
Deep review Sonnet · on-demandFour-dim parallel scoring → Strategic Advisor → full 2-section rewrite.~$0.38 / run
-
Framework gate Sonnet → HaikuSoft-warn on claims without NAO evidence. Never blocks.~$0.01 / check
-
Gap analysis Sonnet · cross-docReads NAO + every document. Returns ranked list of missing evidence.~$0.03 / run
-
School comparison Sonnet · multi-school fitPer-school fit scores, signal match, risk callouts against the NAO.~$0.08 / run
About this prototype
The agent intelligence in this prototype is simulated — scripted deterministic outputs, not real LLM calls. The latency, pipeline orchestration, cost telemetry, and NAO schema are production-accurate. When the real implementation replaces the simulated brains with Anthropic API calls, the UI layer doesn't change.All prototype state lives in your browser's local storage and clears when you click Reset journey in the sidebar.