API — Dhandare Models

What you can ask for

dhandare-4-mini answers defined tasks rather than free-form prompts — that is what makes it deterministic. Pick a task, then send it the way any chat request is sent. There are two ways to name one:

· put task: <id> on its own line — unambiguous, and the recommended form for a programmatic integration; or
· say the action plainly, e.g. "narrate these signals", "reflect on this trade".

GET https://models.quantforge.co.in/v1/tasks returns this same list as JSON — no key required, so you can read it before you subscribe.

Task	What it does	Returns
narrate	Turn a list of scored signals into a per-signal note and one overall read.	strict JSON: {"signals": {name: note}, "overall": "..."}
reflect	Given a closed trade and how each signal was graded, write the outcome and the lesson.	strict JSON: {"outcome": "...", "learning": "..."}
debate	Argue both sides of an already-scored setup, then synthesise.	strict JSON: {"bullCase": "...", "bearCase": "...", "synthesis": "..."}
resume	Re-order and emphasise YOUR OWN resume against a job description. Never invents experience.	a tailored resume plus an explicit GAPS list

Trading-signal narration task: narrate

Turn a list of scored signals into a per-signal note and one overall read.

Your message needs:

· a line `Instrument: <SYMBOL>`
· one or more signal lines shaped `- NAME: bullish|bearish|neutral NN/100`

Example

curl https://models.quantforge.co.in/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dhandare-4-mini",
    "messages": [{"role": "user", "content": "task: narrate\nInstrument: TCS\n- RSI: bullish 61/100\n- MACD: bearish 30/100"}]
  }'

Returns strict JSON: {"signals": {name: note}, "overall": "..."}

Also selected by: narrate, narration, explain these signals, describe these signals, write up these signals, signal commentary

The overall read is weighted by conviction, not by counting rows — five weak bullish signals do not outvote three strong bearish ones.

Post-trade reflection task: reflect

Given a closed trade and how each signal was graded, write the outcome and the lesson.

Your message needs:

· a line `Trade: <SYMBOL> <SIDE> x<QTY>, entry <P>, exit <P>, net P&L <N> (WIN|LOSS|FLAT)`
· signal lines shaped `- NAME: direction NN/100 (graded correct|wrong|neutral)`

Example

curl https://models.quantforge.co.in/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dhandare-4-mini",
    "messages": [{"role": "user", "content": "task: reflect\nTrade: TCS BUY x50, entry 3900, exit 3960, net P&L 3000 (WIN)\n- RSI: bullish 61/100 (graded correct)\n- MACD: bearish 30/100 (graded wrong)"}]
  }'

Returns strict JSON: {"outcome": "...", "learning": "..."}

Also selected by: reflect on, reflection, post-trade review, what did we learn, review this trade, post trade analysis

The per-signal grades are honoured verbatim — this task writes prose, it never re-grades a signal.

Bull-vs-bear debate task: debate

Argue both sides of an already-scored setup, then synthesise.

Your message needs:

· a line `Instrument: <SYMBOL> (<TRADE TYPE>).` — note the bracketed type and the full stop
· a line `Computed score: N/10, <figure> NN%, net direction <dir>.` where <figure> is either `probability` or `direction agreement` — whichever you send is the wording quoted back to you (use `n/a` in place of NN% if you have no figure)
· signal lines shaped `- NAME: direction NN/100`

Example

curl https://models.quantforge.co.in/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dhandare-4-mini",
    "messages": [{"role": "user", "content": "task: debate\nInstrument: NIFTY (Options).\nComputed score: 6.5/10, probability 74%, net direction bullish.\n- RSI: bullish 70/100\n- OI_PCR: bearish 55/100"}]
  }'

Returns strict JSON: {"bullCase": "...", "bearCase": "...", "synthesis": "..."}

Also selected by: bull vs bear, bull and bear, debate, argue both sides, make the bull case, make the bear case

The score arrives already computed and is quoted, never recalculated — this task cannot move a number.

Resume tailoring task: resume

Re-order and emphasise YOUR OWN resume against a job description. Never invents experience.

Your message needs:

· a `JOB DESCRIPTION:` block
· your profile: HEADLINE / SKILLS / EXPERIENCE or PROJECTS sections

Example

curl https://models.quantforge.co.in/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dhandare-4-mini",
    "messages": [{"role": "user", "content": "task: resume\nJOB DESCRIPTION:\nSenior Backend Engineer. Node.js and PostgreSQL required. Kubernetes preferred.\n\nSKILLS: Node.js, PostgreSQL, React\nEXPERIENCE: Built a webhook pipeline handling 1.2M events/day on Node.js + PostgreSQL."}]
  }'

Returns a tailored resume plus an explicit GAPS list

Also selected by: tailor my resume, tailor my cv, resume for this job, match my resume

Requirements you have no evidence for are reported as GAPS, never written in. That is the point: a candidate caught inventing experience in an interview is worse served than one who arrives knowing the gap.

Chat completions — dhandare-4-mini

POST https://models.quantforge.co.in/v1/chat/completions

The standard OpenAI request shape. The message body is one of the tasks above.

The x-api-key header is accepted too, for Anthropic-style clients.

A request that matches no task is answered with an explicit refusal listing the task ids — this model never invents prose to fill a gap. If a task is recognised but the body cannot be read, the reply names the task and shows a working example.

Skills — ready-made features on the real models

POST https://models.quantforge.co.in/v1/skills/<id> · GET https://models.quantforge.co.in/v1/skills (no key required)

One call per feature — no prompt writing. Every answer is JSON that matches the schema below (the model is constrained by a grammar, so it cannot return anything else). Billed at the listed model's token prices. Models start on demand: the first call after a quiet period waits while the model loads (reported as dhandare.engine.cold_start).

Skill	What it does	Model	Inputs
grammar	Corrects grammar, spelling and punctuation without changing the meaning, and lists each change.	dhandare-4-mini	`text`
rewrite	Rewrites text in a chosen tone — formal, friendly, concise or persuasive — keeping every fact.	dhandare-4-mini	`text` `tone`: formal \| friendly \| concise \| persuasive
summarize	A short summary and up to five key points, using only what the text says.	dhandare-4-mini	`text`
translate	Translates between Hindi and English, keeping names, numbers and formatting.	dhandare-3.1-pro	`text` `target`: hindi \| english
extract	Pulls the fields you define (as a JSON Schema) out of free text — invoices, emails, forms.	dhandare-4-mini	`text` `schema`
classify	Picks the best label from the list you give — support tickets, feedback, intents.	dhandare-4-mini	`text` `labels`
resume	Turns a candidate's own details into a structured resume aimed at a job description — no invented facts.	dhandare-3.1-pro	`candidate` `job_description` (optional)

curl https://models.quantforge.co.in/v1/skills/grammar   -H "Authorization: Bearer $DHANDARE_KEY" -H "Content-Type: application/json"   -d '{"input": {"text": "he go to office yesterday and forget his bag"}}'

# → {"skill": "grammar", "model": "dhandare-4-mini",
#    "output": {"corrected": "He went to the office yesterday and forgot his bag.", "changes": [...]},
#    "usage": {...}, "dhandare": {"engine": {"cold_start": false, ...}}}

Task catalogue — JSON

GET https://models.quantforge.co.in/v1/tasks

The same list this page shows, machine-readable, no key required. GET https://models.quantforge.co.in/v1/tasks/<id> returns one task.

curl https://models.quantforge.co.in/v1/tasks

Embeddings — dhandare-embed-1

POST https://embeddings.quantforge.co.in/v1/embeddings

curl https://embeddings.quantforge.co.in/v1/embeddings \
  -H "Authorization: Bearer YOUR_EMBED_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "dhandare-embed-1", "input": "text to embed"}'

Returns 384-dimensional vectors, L2-normalised. input accepts a string or an array of strings.

Usage and billing in the response

"usage": {
  "prompt_tokens": 812,
  "completion_tokens": 143,
  "total_tokens": 955,
  "prompt_tokens_details": { "cached_tokens": 640 }
}

Repeating the same system prompt within 5 minutes bills those tokens at the cheaper cached rate. These are exactly the numbers debited from your wallet.

Errors

Status	Code	Meaning
401	invalid_api_key	Key is wrong or was revoked by regeneration.
401	wrong_product_key	You used the other product's key.
402	insufficient_balance	No credit left for this model — recharge, or take a plan, in the console.
403	account_disabled	Account disabled by an administrator.
404	model_not_found	Unknown `model` id — see `GET /v1/models`.
429	rate_limit_exceeded	Too many requests for this key. Honour the `Retry-After` header.
429	concurrency_limit_exceeded	Too many requests in flight for this key at once.
400	unsupported_dimensions	Embeddings are 384-dimensional and cannot be resized.

A throttled (429) request is never billed — the limit is applied before any metering. Rate limits are enforced per API key.

Client compatibility

Parameter	Behaviour
stream	Supported. `stream:true` returns server-sent events ending in `data: [DONE]`, so the standard SDKs work unchanged. The answer arrives in a single chunk — this engine computes a complete result rather than token-by-token, and chunking it artificially would only add latency.
stream_options.include_usage	Supported — appends the usage-only chunk before `[DONE]`.
temperature, top_p, presence_penalty, frequency_penalty, seed	Accepted and ignored, by design. The engine is deterministic — the same input always yields the same output — so there is no sampling for these to steer. They are accepted so existing client code needs no changes.
max_tokens	Accepted and ignored. Answers are strict JSON for the structured task shapes, and truncating them mid-document would hand you unparseable output — so responses are always returned whole.
model	Validated. Use `dhandare-4-mini`, or pin a build with a dated alias such as `dhandare-4-mini-2026-08`. Responses carry `system_fingerprint`, which changes when the engine build changes.
dimensions (embeddings)	Accepted only at the native width of 384; any other value is a 400 rather than a silently different vector. Size your vector column for 384 dimensions.
encoding_format (embeddings)	Supported: `float` (default) or `base64` (little-endian float32).

What this model is — and is not

dhandare-4-mini is a deterministic work model, not a general LLM. It serves the defined tasks above — trading-signal narration, post-trade reflection, bull-vs-bear debate and resume tailoring — and returns the same output for the same input, every time, with no GPU and no third-party API behind it. Prompts outside those shapes get an explicit "unsupported task" reply: this model never invents prose to look clever.

That is the trade. It will not write you a poem or hold an open-ended conversation. In exchange, the same request gives the same bytes tomorrow — which is what makes it safe to put in a pipeline, and why system_fingerprint is a promise rather than decoration.

API reference

What you can ask for

Trading-signal narration task: narrate

Post-trade reflection task: reflect

Bull-vs-bear debate task: debate

Resume tailoring task: resume

Chat completions — dhandare-4-mini

Skills — ready-made features on the real models

Task catalogue — JSON

Embeddings — dhandare-embed-1

Usage and billing in the response

Errors

Client compatibility

What this model is — and is not