dhandvar

Dhan × Dvar · concept demo

Your backtester, your assistant, your models. Run by Dvar.

Five things we can build and run for Dhan and the Raise group, each with a working demo: a white-label backtester, Dhan Assist, evals on the chatbot, post-training for Artham, and a harness for document scanning.

Start the guided tourTry the backtesterAsk Dhan Assist

~5 s

per daily backtest, in the browser

Same rule, same answer: every run has a hash

246

backtests to verify 10 listed algos

Re-run, later years, nearby settings, higher costs, enough trades

7

checks on every assistant answer

Figures from sources, no advice, confirm before acting, no PII

The five workstreams · about 12 minutes to see them all

Workstream 1Backtester outsourceA managed, verified backtesting service for Dhan, Dhan Algos and Stratzy, plus custom models Dhan presents as its own.In the demo: Type a rule, get a result in seconds; verify 10 listed algos with 246 backtests; see the white-label API and service levels.Week 6: every daily strategy re-verifiedOpen →Workstream 2Customer chatbotDhan Assist, an investing co-pilot that knows the customer's portfolio, explains stocks and F&O, helps with orders and support, in English, Hindi and Hinglish.In the demo: Ask about your portfolio, a stock or an option; get cards and charts, follow-up suggestions, orders drafted for confirmation and advice declined.Week 8: internal pilotOpen →Workstream 3Evals on existing use casesTraces and pass rates on fuzz, ScanX and support traffic, and a gate on every model change.In the demo: A 30-chat scorecard in the browser, and the Eval Console: 180 chats, three models, five checks, one release gate.Week 2: first eval reportOpen →Workstream 4Post-training ArthamTask-tuned Artham versions that beat the current one on Raise's own checks.In the demo: A measured rule-writer eval in your browser, the misses turned into training data, the gate and a model card in Raise's name.Week 14: first tuned releaseOpen →Workstream 5Harnesses for ad-hoc appsOne shared harness for document scanning and similar apps: schemas, checks, review queue, traces.In the demo: Five specimen KYC and income documents (PAN, masked Aadhaar, cancelled cheque, bank statement, contract note): field reads, checks, routing, a review queue.Week 6: document scanning on the harnessOpen →How Dvar helps DhanThe five options on one page: the value to Dhan and what Dvar delivers for each.See the options →The one-page briefWhat each demo proves, the 24-week plan, and what we need from Dhan to start. Prints to a PDF.Open the brief →

One idea runs through all five

Models write words and structure; code does the maths and the checks. The backtester's model writes the rule and the engine writes every number. The assistant answers from tools, and seven code checks run before a reply is shown. A cheaper model ships only when the gate says it is no worse where it matters. That is what lets Dhan put its own name on it.

  1. 1Words in
    A trader's sentence, a support question, a scanned page
  2. 2Small model (Artham Support-1)
    Writes a rule, an answer or a filled schema
  3. 3Code checks
    Schema, figures from sources, no advice, formats, totals
  4. 4Router
    Passes → ship it. Fails → large model or a person
  5. 5Traces and evals
    Every run logged; misses become the next training set
Words used in this demo
Frontier model
A large hosted model from a company such as Anthropic, Google or OpenAI. Strong, and the most expensive per answer.
Small tuned model
A 7B model like Artham, trained further on Raise's own tasks. Much cheaper, and good enough once it passes the gate.
Run hash
A fingerprint of the rule plus the engine, data and charges versions. Same hash, same result, any time.
Release gate
The rule a new model must pass before it takes traffic: no worse than the current one on any critical check.
White-label
Dhan's name on the product and the models; Dvar builds and runs them in the background.
Harness
The shared plumbing around a model: schemas, checks, routing, review queue, traces.

Everything here is synthetic and built by Dvar; no Dhan data was used. The full project plan is shared alongside this demo.