Self-hosted statutory research, deterministic tax and GST tooling, and document intelligence — architected so confidential client data never leaves your infrastructure and never reaches a third-party AI provider.
No credit card required · Free plan available · SEBI-compliant
FinRace AI is a sovereign, self-hosted AI platform for Indian chartered accountants, tax advocates and law firms. It runs entirely on infrastructure you control, with no external LLM API calls — combining deterministic (non-LLM) tax and GST computation, citation-validated statutory-provision research, and document intelligence. It is built for on-premise and air-gapped deployment and follows a DPDP-aware design: per-tenant data isolation, an immutable audit log, a consent ledger, and configurable retention.
No LLM in the number path. Every research citation is validated before it renders. Gated capabilities are labelled, not implied.
Old vs New regime comparison (Section 115BAC), versioned by assessment year. Advance-tax installment schedule with Section 234B / 234C interest. Real statutory math — no LLM in the calculation.
GSTR-2B vs purchase-register ITC reconciliation with Section 17(5) ineligible-credit flagging, and a draft ASMT-10 reply from the mismatch data. GSTR-1 / 3B / 9 consistency, reverse-charge flagging, e-invoice IRN validation.
Section rate and threshold lookup against the current FY table; deterministic TDS-payable computation; reconciliation of claimed TDS credits against Form 26AS / AIS.
Explains the likely notice type behind an AIS / TIS / 26AS discrepancy. Diagnostic only — it does not auto-draft a reply to an income-tax notice.
Retrieval over ingested statute text and CBDT / RBI / SEBI / MCA circulars, with every citation checked against retrieved context or a canonical statute list before it renders. No case-law or tribunal corpus — statutory-provision and circular research, not precedent search.
Legal notices, petitions, replies and affidavits in court format from the facts you provide, as editable .docx. CA working papers and audit engagement letters from real computed figures — each labelled “AI-generated, requires professional review”.
Companies Act compliance calendar with real AGM / board-meeting / AOC-4 / MGT-7 windows, Section 188 related-party-transaction evaluation, MCA e-Form pre-audit, and deterministic EGM-timeline / buyback / ESOP-vesting calculators. DIN validation is format-only, not a live MCA status lookup.
Point-in-time fundamental snapshots — ratios, quality / growth / risk scoring, valuation verdicts — from real financial data, with OCR and field extraction for statements you upload.
Concentration and correlation risk analysis across your holdings, position-sizing checks, and technical (daily / intraday-where-available) and fundamental snapshots — ratios, quality / growth / risk scoring, valuation verdicts — for Indian equities. Options Greeks and a strategy builder for derivatives.
Credit Appraisal Memos built from real GSTR-3B, ITR and bank-statement data — DSCR, EMI / FOIR / LTV computed deterministically, no mocked figures. An Early Warning System for EMI bounces, turnover variance and circular-trading patterns. Proposed lending actions checked against real RBI/SEBI circulars with citation validation. No Basel III capital-adequacy scoring or AML/CFT watchlist screening — those require data subscriptions FinRace AI does not bundle.
More for CAs and tax consultants on the chartered accountants page, and for law firms on the legal page.
Six professional-vertical LoRA experts — one sovereign, self-hosted base model, fine-tuned per persona — each scoped to what's actually working today. Capabilities still pending real statutory source data are labeled, not implied.
The six personas share one calibrated intelligence stack, one document pipeline, and one tenancy & audit model — plus two adjacent surfaces for the same firms.
A seventh persona beyond the financial six — for talent teams inside the same firms.
Deterministic document handling for onboarding desks.
Shared by every vertical, not bolted on per persona.
Auditable by construction.
Schema-Compliant Tally Prime XML File Generator for secure, batch-importable ledger operations — no cloud-to-LAN connection required.
BIAN-standard event webhook receiver for transaction monitoring, wired to the real Early Warning System when transaction data is present.
Per-tenant SSO, your own logo and brand colours, usage-metered billing, and scoped API keys with rate limits. Row-Level Security isolation, auth-boundary-only security logging, no third-party analytics or cross-site tracking anywhere.
Click below to preview how our terminal integrates financial AI with raw analytical data.
Analytical answers are structured, not free-form: Observational Evidence and Cross-Signal Synthesis are collapsible so the canvas stays uncluttered, Statutory/Regulatory Grounding runs every citation against retrieved context and a canonical whitelist before it's shown, and the final Determination is always visible, never buried behind a toggle.
Ask questions in plain English, upload bank statements, financial documents, or query stock symbols.
Fast API router directs deterministic math/technical tasks to local compiled engines, and complex reasoning to the sovereign Multi-LoRA model — 100% self-hosted, nothing ever leaves our infrastructure.
Answers arrive alongside interactive widgets (tax tables, chart screens, python consoles) directly inside the interface.
A separate, versioned partner API — not an afterthought bolted onto the consumer app. Built for banks, NBFCs, and fintechs who need our analysis engine inside their own product.
Every request needs either an API key or a session JWT — the same auth used across the whole platform, nothing partner-specific to relearn.
Register a webhook URL and secret in your organization settings to receive HMAC-signed event payloads the moment an async research job completes — no polling required.
Analysis
Full decision + prediction + risk assessment for a symbol
Same engine, with optional natural-language query context
Explainability trace only — the reasoning behind a verdict
Knowledge Base
Ingest a PDF/Excel/CSV into your tenant's document index
Semantic search across your uploaded documents
Tax & GST Tools
Old vs. New regime tax liability comparison
Reconcile a purchase register against GSTR-2B entries
Typed client libraries generated directly from the API's OpenAPI spec — no drift between docs and code.
The partner API ships its own always-on Swagger UI, ReDoc, and raw OpenAPI schema — separate from the main app's internal (debug-only) docs.
Partner API access is included from the Institutional plan, with full webhook + team-seat support. Every response carries the same SEBI-compliance disclaimer as the terminal — this is informational output, not investment advice, whether a human or your system reads it.
FinRace AI is a sovereign, self-hosted AI platform for Indian CA firms, tax advocates and law firms. It runs on your own infrastructure with no external LLM API calls. For a CA firm it provides deterministic (non-LLM) tax and GST computation — Old vs New regime, advance tax with 234B/234C interest, TDS rate lookup and computation, GSTR-2B ITC reconciliation with Section 17(5) flagging, GSTR-1/3B/9 consistency checks — plus citation-validated statutory-provision and circular research, fundamental/balance-sheet analysis, and .docx drafting of CA working papers and engagement letters. Every research answer arrives with the sources it was grounded in.
Yes — that is the core design. Reasoning runs on a self-hosted model on infrastructure you control; no client data is sent to OpenAI, Anthropic, Google or any third-party AI service, and there are no analytics or tracking scripts in the product. It is built to run on-premise or air-gapped, with per-tenant data isolation, an immutable audit log and configurable retention.
FinRace AI runs deterministic GST checks with no LLM in the calculation path: GSTR-2B vs purchase-register ITC reconciliation with Section 17(5) ineligible-credit flagging, GSTR-1/3B/9 consistency, reverse-charge liability flagging, and e-invoice IRN validation. From a reconciliation it can produce a draft ASMT-10 reply. It does not search GST case law or a GST circular database — the GST feature set is computation and reconciliation, not precedent research.
Partly. It analyses an AIS / TIS / 26AS discrepancy and explains the likely notice type behind it — diagnostic, not a drafted reply. On the GST side it can draft an ASMT-10 reply from reconciliation data. It does not auto-draft replies to income-tax department notices.
Yes. It drafts 45 court- and registration-formatted document types across 10 practice areas — legal notices (including Section 138 cheque-bounce notices), plaints, written statements, applications, reply affidavits, writ petitions, NDAs and other commercial agreements, sale / lease / gift / mortgage / partition deeds, general and special powers of attorney, divorce / maintenance / custody petitions, and wills. Each type asks only for the fields its court or statute requires, produces an editable .docx, and reproduces the facts you enter verbatim rather than re-writing them. Drafts are AI-generated and require advocate review and signature before filing; FinRace AI does not e-file, e-register or serve documents.
No. There is no case-law, judgment or tribunal-precedent corpus. Legal research in FinRace AI is statutory-provision and circular research — retrieval over statute text and CBDT / RBI / SEBI / MCA circulars, with every citation validated against the retrieved context or a canonical statute list before it renders.
They describe the architecture and the supported deployment model, not an unconditional guarantee. “Sovereign” means all inference is self-hosted with no third-party AI calls. The platform is built for air-gapped deployment — it can run without outbound internet, and a hardened mode disables non-essential external calls. The reference cloud deployment is internet-connected; air-gapped is a supported install target. Data handling is DPDP-aware by design (consent ledger, immutable audit log, configurable retention) — an architectural posture, not a certification.
(1) All inference self-hosted — no data to external AI providers. (2) Deterministic math for anything statutory (tax, GST, TDS) rather than an LLM guessing numbers. (3) Citation validation on research output. (4) Honest scoping — the tool should tell you what it does not do (no case-law corpus, no Form 3CD) rather than implying it. (5) Per-tenant isolation, an audit log, and retention you control. FinRace AI is built to each of these.
Options analytics, bank statement/GSTN reconciliation, statutory research, and document integrity checks — all running on self-hosted infrastructure, nothing sent to a third-party AI provider.
Talk to us about deploying FinRace AI inside your bank, law firm, or practice.