v2.0 · Updated 2026-05-13
Presenter mode · Synapi
—
01 / 23
00:00
Speaker notes will appear here.
Up next
—
Use ← → or Space to advance both windows. Closing this window keeps the deck running.
Speaker notes
—
Q&A Appendix

For investor follow-up

1 / 3
A1 · Risks & mitigations
Honest assessment

What could go wrong — and how we're handling it.

Users let AI answer for themTimed recall · explain-back · variation across reps · uncertainty flags. We measure what survives without the model.
Scoring overconfidenceBayesian estimation with explicit confidence intervals. "Needs more evidence" states. Recency-weighted.
Proof shared out of contextConsent-based Proof Cards show scope, recency, evidence type, limitations. Defensible if questioned.
Model dependency / lock-inModel-agnostic by design via Lovable AI Gateway. We route across Gemini, GPT-5, frontier previews.
Incumbents copy the wedgeMoat is time-series mastery data + Learner DNA. They can copy listings, not a year of trusted learning evidence.
Cold-start on institutionsB2C-first sequencing: own the learner data before selling proof to schools and employers.
A2 · Unit economics
Numbers

Healthy consumer SaaS economics.

Directional Consumer Pro economics at $20/mo. Marketplace and institutional layers stack on top.

ARPU
$240
/yr · Consumer Pro
Gross margin
~78%
after model + infra
CAC target
<$40
organic + Proof Card share
LTV / CAC
~6×
at 18mo retention
Multi-line stack

Consumer Pro is the wedge. Proof Packs ($49 one-time) and cohort dashboards ($2K–10K/yr) sit on top of the same data, so contribution margin grows. Employer AI-literacy contracts ($50K–250K) and eventual marketplace take-rate are step-functions on the existing user base, not new acquisition.

Numbers are directional. Hard data lands in the next 8–12 weeks from the closed beta.

Table of contents

Jump to a slide

S
Synapi — Investor Deck
01 / 20

Navigate the deck

→ Space next slide

← previous slide

Home first   End last

T table of contents

N notes drawer (solo review — same screen)

M presenter mode (live pitching — opens a 2nd window)

A Q&A appendix (risks · unit economics)

F fullscreen   P print/PDF

1–9 jump to slide   Esc close overlays

Live pitching workflow: Press M to open Presenter window. Drag that window to your laptop screen, then mirror/share the main deck window to the projector or call. Both stay in sync as you navigate.

Most AI is built to do your thinking.
Synapi
We're building one that keeps the thinking yours.
The metacognitive OS for the AI era · Use AI. Don't become it.
Investor Deck · 2026
02 · Problem
The illusion

The tools making you feel sharpest are quietly making you dullest.

Randomized controlled trial · PNAS 2026
~1,000 students: a GPT-4 tutor vs. no AI
No AI With AI tutor
no-AI baseline = 100 100 148 +48% 100 83 −17% same students — score collapses on the real test Practice problems WITH the AI tutor Real exam NO AI · weeks later

Confidence soared. Actual ability dropped. The students never knew the difference.

03 · The blind spot
The gap no one sees

And nothing — anywhere — reveals it in time.

58%
APA · 2026 · n=1,923
of workers say AI "did most of the thinking" — and report weaker ownership and shallower thought.
83%
MIT Media Lab · 2026
of AI-assisted writers could not recall a single line of the essay they had just submitted.
+4 pts
Fernandes et al. · 2024
AI users overrate their own performance — confidence stops tracking what they actually know.
Products
AI tools measure speed and output volume. None measure whether the user actually understood what was produced.
Schools
Institutions grade the submission, not the cognition behind it. AI-polished work is indistinguishable from mastery.
Workplaces
Employers measure throughput and deadlines. They have no signal for whether a team member's skills are growing or atrophying.

Speed without measurement becomes capability debt — invisible until it's irreversible. The market need: an instrument that proves human capability after AI assistance.

04 · Solution
What we've built

A working metacognitive OS.

Five specialized agents orchestrated in parallel, sharing state through a Personal Knowledge Graph and pgvector RAG layer. Goal → personalized skill tree → Socratic practice → Bayesian mastery → portable Proof Card. Shipped on iOS and Web today.

synapi · goal Live
New goal
What do you want to master?
Land an ML engineer role in 12 weeks
Add context · optional
PDF
Notes
Link
Course
Generate skill tree
1Capture intent
Horizon agent
synapi · map Synced
Skill tree · 42 nodes 14d
8 of 42 mastered+3 this week
Linear Algebra Prob & Stats Calculus ML Basics Neural Nets SGD Optim. Backprop ← now
Start practice
2Generate map
Horizon + Research
synapi · practice Recording
Retrieval 00:42
Backpropagation
Explain in your own words: why does backprop require a differentiable loss function?
Your answer · no AI hint
Because gradient descent needs a gradient, and a gradient only exists if the loss is
Skip
Submit
Rep 3 of 8
3Practice
Reflect + Diagnostic
synapi · proof Verified
Mastery trajectory
Overall
84/100
+12 · 14d
Reps
347
28 this wk
Retrieval92
Compression87
Synthesis81
Judgment76
Generate Proof Card
4Prove + share
Analyst + Learner DNA
Agents Horizon Reflect Analyst Diagnostic Research
Infra Personal Knowledge Graph pgvector RAG SM-2 spacing Bayesian mastery Learner DNA
Model-agnostic Gemini GPT-5 Frontier preview via Lovable AI Gateway
iOS + Web · working today
05 · Roadmap
Where we're going next

From working OS to true metacognitive infrastructure.

Full multi-agent orchestration

Planner–executor loop: a top-level Coach agent delegates to Curriculum, Diagnostic, Retention, Research, and Coach-of-Coach sub-agents via MCP, negotiates trade-offs, and produces one coherent learning plan per turn — with visible reasoning traces.

Deep multimodal ingestion

Lecture videos and long-form audio transcribed and decomposed into skill-tree branches, with timestamped retrieval back to source moments.

Adaptive assessment v2

Activation challenges that probe transfer, not just recall. Auto-recalibration when a learner outperforms or struggles against predicted mastery.

Calendar + LMS via MCP

Spaced-practice sessions written directly into Google Calendar. Skill trees connected to Notion, Canvas, Linear coursework and sprint goals.

North star: a portable Learner DNA — an exportable cognitive profile that travels with you across goals, domains, and life stages.
06 · Why it works
The science

50 years of learning science, productized.

Tutors and AI explainers optimize for comprehension in the moment. Synapi optimizes for what survives without the model — recall, transfer, judgment, proof over time.

Active recall
Roediger & Karpicke · 2006
The testing effect: retrieval practice produces 2–3× better long-term retention than re-reading. Synapi forces retrieval on every rep.
Spaced repetition
Ebbinghaus 1885 · SM-2 (Wozniak)
Reviewing at increasing intervals defeats the forgetting curve. Synapi runs SM-2 per concept, with intervals adapting to real demonstrated recall.
Bayesian Knowledge Tracing
Corbett & Anderson · 1995
Probabilistic skill estimation with explicit confidence intervals. Foundation of every serious tutoring system since.
Desirable difficulty
Bjork · 1994
Harder retrieval at the edge of ability builds stronger long-term memory. We deliberately interleave and vary contexts.
Springer 2026 (N=912): offloading to AI helps learning — but only when paired with vigilance. Synapi is the vigilance layer.
07 · Market
The opportunity

$580B converging on the proof layer.

Feb 2026
EU AI Act Article 4 enforceable. Every employer in the EU needs AI-literacy proof. Today: nobody has it.

High-intent learners first (Consumer Pro), then cohorts and employers (B2B2C), then institutions buying verified capability.

TAM
$580B
SAM
$52B
SOM
$1.2B
TAM · Total addressable
$580B

Global learning ($400B by 2030) + corporate reskilling ($150B) + hiring & verification tech ($30B).

SAM · Serviceable
$52B

AI-era learners + AI literacy compliance (EU AI Act Art. 4) + cohort/bootcamp evidence + AI-aware screening.

SOM · 3-yr obtainable
$1.2B

Consumer Pro ($20/mo × 2M high-intent learners) + Proof Packs + early cohort dashboards.

Reaching $100M ARR requires ~0.2% of SAM. Category-creating, not market-expanding.

08 · Traction
Traction

Already validated.

2nd Place
Hackathon · 2026
Love at Scale Hackathon
Wilson Sonsini Goodrich & Rosati·Palo Alto·2nd place
Built in 6 weeks 5 agents in production 60s skill-tree generation iOS + Web shipped
Public demand
1,200+
Rednote likes · 150+ beta requests
On the product
115
Active users
First revenue
1
Paying customer · annual plan
This may be the greatest thing since sliced bread. I tried using Synapi to help me understand a research paper. Thank you for making this!
AI Engineer · ClawCamp
I found real value in the dialog and questions. I am a Socrates fan.
Frank C.UC Berkeley · Legal Studies
I can imagine college students becoming obsessed with this — constantly coming back to check their skill growth.
YunanEarly User
The visualizations at the end blew me away — seeing my knowledge mapped out was incredibly motivating!
Y.Early User
09 · Business model
How we make money

Revenue starts before the marketplace.

Five revenue lines. First buyer: the learner who feels the pain. Larger buyer: the institution that must trust, report, hire, train, or certify.

Phase 1
Consumer Pro
Ambitious learners & builders
$20/mo
Phase 2
Proof Packs
Job seekers, students, creators
$49 one-time
Phase 3
Cohort dashboards
Bootcamps, creator schools
$2–10K/yr
Phase 4
Employer AI literacy
Teams adopting AI · EU AI Act
$50–250K contract
Phase 5
Marketplace
Buyers of verified expertise
Take-rate later
The enterprise wedge: companies already spend billions on Copilot, ChatGPT Enterprise, and internal agents — but seat adoption ≠ capability gains. Synapi raises the metacognitive floor — the ROI multiplier on AI tools they already pay for.
10 · Team
Why us

The founder is the user.

Cognitive science × technical depth × capital fluency — built by the person who lived the problem.

Yiya WangYW
Yiya Wang
Founder & CEO
Cognitive Science, UC Berkeley. Built Synapi's product loop from her own daily workflow with Claude and Anki — she is the user we build for. Founded Berkeley EdTech; product at Imagi (edtech, Stockholm). Hackathon winner.
UC Berkeley · CogSciBerkeley EdTechex-ImagiHackathon winner
Angelina RichterAR
Angelina Richter
Co-founder & CTO
Math & CS, Stanford. Owns Synapi's mastery scoring and AI infrastructure. Currently SWE at Rippling; previously engineering at NASA; investor at Mana Ventures — builds at scale, reads the market.
Stanford · Math + CSRippling SWEex-NASAMana Ventures
"
After a week of learning with Claude, I couldn't remember what it had just explained. The artifact was perfect. My memory was empty. That moment was Synapi.
Yiya Wang · Founder · UC Berkeley, finals week 2025
The horizon

Use AI.
Don't become it.

AI makes everyone faster
↓
Speed becomes common
↓
Human capability becomes scarce
↓
Proof becomes valuable
↓
Synapi

Where goals become mastery, and mastery becomes trusted opportunity.

We're early. We're moving fast. We're raising a seed round.
hello@coaur.com Yiya Wang · Angelina Richter