Step 01
Step 1
Type a question; the chat tab routes it to the FinAi model running on our self-hosted vLLM cluster.
Chat & VoiceRetail· marketing-claim · courtroom-three-roles
Conversational AI advisor.
Ask anything about your money — every meaningful answer is argued before it reaches you.
The problem
Why this exists.
Most finance chatbots are either FAQ lookups dressed as AI or unreviewed LLM outputs that cite nothing.
We want an advisor that admits when it does not know, cites when it claims a number, and gets argued by an internal Skeptic before the answer ships.
How it works · 3 steps
From input to outcome.
- Step 02
Step 2
Every meaningful output passes through the Courtroom — Proposer, Defender, Judge — and the HallucinationCheck grounding pass.
- Step 03
Step 3
Answers cite sources and admit uncertainty in the same sentence.
Under the hood
The models and routes behind it.
Models called
API routes
- POST /api/v1/chat
Compliance posture
- Foundation models are Apache 2.0 or MIT — no Claude / GPT / Gemini in the inference path.
- Every consequential output is signed by the Courtroom under courtroom-replayable-audits.
- DPDP §6 — chat turns are captured per the consent flags you set.
Interactive demo · mocked at the edge
See it work.
Ask Saraswati (mock)
Loading the demo…
What it's not
The advisor is assistive, not authoritative. The Courtroom + HallucinationCheck reduce error rate but do not eliminate it — we lead with 'argued before it reaches you', not with absolute correctness.
Related · features & stories