Dhan × Dvar

How Dvar helps Dhan

Five ways to work together. Take one or all five: each stands on its own, and they share one way of working. Models write words and structure, code does the maths and the checks, and nothing ships without passing a gate.

Option 1

Backtester outsourcing and white-label models

Value to Dhan

Launch and scale backtesting without building a quant-infrastructure team
Every listed algo re-verified, so a strategy page never shows a number that can't be reproduced
Models and API under Dhan's name: Dhan owns the product, Dvar runs it

What Dvar delivers

The backtest engine and white-label API, with run hashes and service levels
A rule-writer model (Artham Rule-1) that turns plain words into rules
Verification of every listed strategy, model cards in Raise's name

First result: week 6, every daily strategy re-verified

See it in the demo →

Option 2

Broker chatbot

Value to Dhan

Support questions answered in seconds, in the customer's language, with tickets only for what needs a person
A GR1-class experience: portfolio answers, charts and payoffs, orders drafted for review
Compliant by construction: no advice, no order placed without confirmation, no PII shown

What Dvar delivers

Dhan Assist on Dhan's account, order and help-centre tools
A small tuned model first, the frontier model only on a critical miss
Runtime checks on every reply, hand-off to agents, full traces

First result: week 8, internal pilot

See it in the demo →

Option 3

Evals

Value to Dhan

Know a model change is safe before customers see it
Move traffic to cheaper models with evidence, not hope
An audit trail of what each model said and how it was checked

What Dvar delivers

The Dvar Eval Console on your traffic: gold sets, rubrics, a calibrated judge
Critical checks for advice, fake confirmations and PII leaks
A release gate on every prompt or model change

First result: week 2, first eval report

See it in the demo →

Option 4

Post-training models like Artham

Value to Dhan

Own models that do Raise's tasks better than a general model, at a fraction of the cost
Data and weights stay with Raise, hosted in India
Each release proven on Raise's own checks before it ships

What Dvar delivers

Training sets built from traces and review queues
Fine-tuning runs, evals and gated releases
Model cards in Raise's name; hosting on Dvar Inference or Raise's cloud

First result: week 14, first tuned release

See it in the demo →

Option 5

Harnesses for apps like document scanning

Value to Dhan

New AI apps in weeks, each with the same checks, not its own plumbing
Fewer manual reviews: only the fields that need a person reach one
Masking and retention that fit DPDP, the same in every app

What Dvar delivers

Schemas, format and cross-document checks, routing small-to-large
A review queue whose decisions become eval and training data
Document scanning first, on the harness, with traces and evals

First result: week 6, document scanning on the harness

See it in the demo →

Under your name

Dhan's brand on the product and the models; Dvar builds and runs them in the background.

Measured, not promised

Every model is scored on your own checks before it takes traffic, and again on every change.

Your data stays yours

India hosting or Raise's own cloud, personal data masked in every trace, DPDP-ready retention.

A sensible start

Evals (option 3) and the backtester (option 1) first: they need the least from Dhan and give the first results in weeks 2 and 6. Dhan Assist and post-training build on the eval sets and traces they create; the harness takes document scanning live on the same plumbing.

Concept demo by Dvar. Timelines are estimates before seeing Dhan's data, stack or volumes. The full brief →