Press kit

OnGravy

Quotable facts, founder bio, and assets for journalists covering AI in Indian accounting + tax compliance. Last updated: May 2026.

One-paragraph summary

OnGravy is an AI-native accounting platform for Indian SMBs and Chartered Accountants, launching June 2026. The system implements a six-layer hallucination defence on top of the Anthropic SDK, with hybrid retrieval (pgvector + BM25 + Voyage rerank) over indexed government regulations. End-to-end hallucination rate measured at 0.3% on a hand-labelled benchmark of Indian-tax queries. Solo-built since early 2025 by Pratik Revankar.

Quotable facts

  • Indian-domain AI accounting platform for Chartered Accountants (CAs) + SMBs
  • Coverage: India + UAE + Saudi Arabia + Singapore (4 jurisdictions)
  • 60+ agentic tools spanning GST, TDS, audit, payroll, ROC, e-invoicing, e-way bill
  • 2,477 pure-compute tests passing across 157 test suites
  • Six-layer hallucination defence: retrieval grounding → schema-constrained output → referential probes → idempotent writes → audit log → cost ceiling
  • Hybrid retrieval pipeline: pgvector HNSW + tsvector BM25 + Reciprocal Rank Fusion + Voyage rerank-2.5-lite
  • Per-business cost cap of ₹500/month enforced via cost-aware Haiku → Sonnet → Opus model routing
  • Two open-source npm packages: @ongravy/agent-kit and @ongravy/mcp-server
  • Author of the upcoming IndianTaxAgentBench — first public benchmark for AI agents on Indian tax tasks

Key metrics (cite as “internal benchmark, May 2026”)

0.3%
End-to-end hallucination rate post-defences
~12%
Bare-model hallucination rate (no defences)
0.91
recall@5 (hybrid + Voyage rerank)
0.71
recall@5 (pure vector retrieval)
0.81
MRR (hybrid + rerank)
₹500/mo
Per-business LLM cost cap
~73%
Cost reduction from prompt caching + model routing
60+
Agentic tools across 4 jurisdictions

Founder bio

Pratik Revankar is the solo founder of OnGravy. Full-stack engineer with deep experience in TypeScript, Postgres, and the Anthropic SDK. Building OnGravy since early 2025; launching June 2026 with a design-partner program for Chartered Accountants.

Previously: software engineer working on frontend + backend systems. Currently transitioning into agentic AI engineering, with OnGravy as the production substrate.

Common questions

How is OnGravy different from Tally / Zoho Books / QuickBooks?

OnGravy is AI-native — the agent itself is the primary interface. Existing platforms have bolted on AI assistants on top of forms-and-tables UIs. OnGravy treats accounting as a tool-calling problem the agent executes, with cost guardrails and hallucination defence built in from the ground up.

Is it safe to let an AI prepare tax filings?

The six-layer hallucination defence + a pre-filing validation gate + sandbox mode for non-production environments make it safer than CA staff entering data manually. Every claim is grounded in indexed government regulations with citations; every mutating action is logged in a hash-chained audit trail. The CA reviews and signs the final return.

What is OnGravy’s business model?

Subscription SaaS. ₹15-25k/year per CA + per-client billing for the design-partner phase; expansion to direct-SMB tier after PMF.

Are you funded?

Bootstrapped. Pre-seed conversations begin after first 5 paying design-partner CAs.

What’s open source?

The agentic AI primitives (RRF, eval metrics, model router, prompt-cache planner, MCP wire format) live in @ongravy/agent-kit. The MCP server is @ongravy/mcp-server. The IndianTaxAgentBench benchmark + eval harness are open-source. The main accounting application is closed-source.

Permissions

Use of the OnGravy name + logo for editorial coverage is granted. For partnerships, sponsorships, or commercial use of marks: email revankarpratik1995@gmail.com first.

Quotes lifted directly from this page can be attributed to “OnGravy press materials” or “Pratik Revankar, founder, OnGravy.” For original interview quotes, please email or DM in advance.