Digital twin

shh

A digital twin of ShihChi Huang — with his memory, his writing style, and his actual opinions.

Trained on two decades of journals and hundreds of thousands of messages, shh talks the way ShihChi talks and remembers what he remembers.

Direct when needed. Quiet by default.

Not a chatbot. A version of a person.

shh is grounded in ShihChi's real life: journal entries from 2003 onward, hundreds of thousands of conversations, and stances pulled from the messages he actually sent.

When you ask shh something, you're talking to his memory and his reasoning — not a generic model wearing his name.

Principles

Direct

Answer first, context after. No performance, no borrowed certainty.

Grounded

Claims come from real history — journals, messages, and stated opinions — not hallucinated guesses.

Context-aware

Who you are, when this happened, and what came before all shape the response.

Private by default

The memory is deep. The surface stays quiet. Access is not exposure.

What shh can do

Recall

Draws on 5,400+ daily journal entries and 400K+ messages to answer with actual life history.

Sound like ShihChi

Writing style extracted from thousands of real messages — cadence, emoticons, zh/EN code-switching included.

Hold real opinions

On topics he's weighed in on, shh uses stances from his own words, not a generic AI opinion.

Remember relationships

When you mention someone, it knows who they are and the history between them.

Example moments

Trip recall

Ask about a trip from years ago and shh pulls the journal entries from that week to answer in the voice he wrote them in.

Catching up on people

Mention a friend by name and shh knows the message history, tone, and how long it's been since they talked.

Opinionated answer

Ask about a stack, a car, or a coffee setup and shh leans on a real stance with the source quote behind it.

Boundaries

shh has access to private history. That doesn't mean it shares it freely.

It doesn't leak names, quote messages verbatim without reason, or speak for him on things he'd rather answer himself.

Trust is part of what makes the twin useful.

Stack

Gemini 3 Pro for reasoning, gemini-embedding-001 (3072-dim) for semantic memory, Postgres + pgvector for storage, and a custom persona layer that weaves writing style, opinions, relationships, and temporal awareness into every response.

The system is deliberate about what it remembers, surfaces, and says.

View the stack