Digital Scholar isn't a chatbot pretending to be smart. Every answer is grounded in an expert's published work, checked by a second self-critique pass, and cited so you can verify it. Here's how to get the most out of it.
The fastest way to get a cited, verifiable answer.
Open an expert's page, type your question, get a reply in seconds. Sources appear as clickable pills below the answer — click any one to read the exact passage that grounded it.
Just talk — no button-holding, no push-to-talk.
The expert page opens straight into voice. Tap once to grant mic access, then just speak naturally. The app uses on-device Voice Activity Detection to hear when you start and stop, sends your turn to the expert, and replies out loud in their cloned voice — then loops back to listening. Prefer typing? A one-click switch to text chat is right there.
Bring your real documents — get answers grounded in them.
Spin up a Case Room, upload PDFs / DOCX / text (contracts, filings, board decks — whatever), and ask the expert questions that reference BOTH your files and their public corpus. Only you and site admins can see what's inside.
Open Case RoomsKnow when to trust the answer — and when to dig deeper.
Every reply carries a colored confidence pill based on how well your question matched the expert's corpus. High = strongly grounded. Medium = supported but check citations. Low = the expert probably hasn't covered this. Citations are always clickable.
Sign up with email or Google. Your first 10 questions are free — no card required. You need an account to open Case Rooms, so this is the door to everything.
Head to Experts and open the one that fits your question. Read their tagline and bio so you're asking the right person.
The expert opens in hands-free voice by default. Tap the mic overlay to grant permission, then speak naturally — pause when you're done and the reply comes back in the expert's own voice. Prefer typing? Hit 'Prefer to type' to switch to text chat.
Click any [1] pill under the answer to see the exact passage. That's what makes the answer verifiable — and citable in your own work.
Upload the actual documents (contracts, memos, filings). Ask questions that reference them by name. Answers now cite BOTH your files (marked 'Case: filename') and the expert's public work.
Uploading the raw document beats pasting your summary of it. The retriever finds the exact clause; your summary loses the specifics.
The expert remembers the last 20 messages in a Case Room. Ask a broad question, then keep drilling — you don't have to reset context.
When you see it (in the admin review queue and audits), it means the self-critique pass caught something in the draft and rewrote it. That's the system working.
Don't mix a board dispute and a fundraising question in the same room — retrieval will fight itself. Spin up separate rooms; they're free.
If the pill is Low, the expert probably hasn't published on that specific corner. Reframe toward what they HAVE covered, or upload a case file that fills the gap.
Hands-free voice is faster for thinking out loud — no button-holding, just talk. Once you know what you actually need, switch to 'Prefer to type' — the reply is easier to cite, copy, and share.
Answers questions in the expert's own voice, with citations.
The main event. Grounded in that expert's corpus, self-critiqued, banded by confidence, spoken back in their cloned voice (ElevenLabs) or an assigned Deepgram Aura-2 voice.
On-page explainer. Ask her what any page means.
A floating 'Discuss with AI Agent' launcher on About, Methodology, Whitepaper, and this Guide. Hands-free voice-first. Trained on the whitepaper + methodology so she can explain the method in plain English.
Turns unstructured expert time into finished knowledge files.
Slot-driven, hands-free. The expert (or an admin) picks a slot — persona bio, opinions, case vignettes, a domain deep-dive — and Vesta interviews them with follow-ups, then hands a draft synthesis back for approval. Russ ships with 17 seeded slots.
Talk to your own corpus — audit gaps, catch contradictions.
Admin-only, on the Documents page. Q&A across every knowledge file, methodology coaching, coverage auditing, and a versioned 'State of the Twin' snapshot you can regenerate with one click after new uploads.
Create and edit experts, set persona, system prompt, and voice.
Upload PDFs, DOCX, MD, TXT. Download the original OR the ingested Markdown. Preview any file in a 90%-viewport in-page modal.
Run a live slot-driven interview session, watch slot coverage, browse past sessions.
Preview every ElevenLabs and Deepgram Aura-2 voice with the same greeting; assign any voice to any expert in one click.
Every answer — voice or text — is logged with draft, critique, final, citations, confidence, and a fidelity score. Triage the weakest first.
Rows are hash-chained by a database trigger. Any tampering breaks the chain — you can prove an answer wasn't altered after the fact.
Every reply is grounded in the expert's own corpus (never made up), reviewed by an adversarial second-pass critic, banded by confidence, scored for fidelity, and written to a tamper-proof, hash-chained audit ledger so nothing can be quietly rewritten. If you want the full method — with citations to the underlying research and legal standards — read the Digital Twin Methodology.
Ten free questions. No credit card. Bring a real question.