What sectors does Vihaya serve?

Vihaya is sector-agnostic by design. The same engine serves healthcare payors, banks and NBFCs, insurers, telecoms, manufacturers, and public-sector workflows. Each engagement deploys the engine into one regulated enterprise’s tenant with a sector-specific decision layer on top.

What does a Vihaya pilot cost?

Pilots are fixed-price 12-week engagements, quoted after a 30-minute discovery call. Vihaya is pre-revenue; design-partner customers get materially better economics in exchange for early production deployment and roadmap feedback.

How long is a Vihaya pilot?

A Vihaya pilot runs for 12 weeks from discovery to handoff. Phases: weeks 1–2 access and policy ingest, weeks 3–8 integration and eval-set build, weeks 9–11 shadow run, week 12 handoff with runbook and reviewer training.

Where does the Vihaya system run?

The system runs inside the customer’s cloud tenant and their region. Vihaya embeds into the customer stack rather than hosting it as SaaS. Data stays in the customer environment with zero SaaS egress.

How does confidence-floor escalation work?

Every Vihaya decision carries a numeric confidence score. Cases below the configured threshold route to the customer’s existing reviewers with the agent’s recommendation, rationale, and citations attached. The reviewer’s override signal trains the adaptive guardrails.

Which LLMs does Vihaya use?

Vihaya is model-agnostic. The default is OpenAI GPT-4o. Anthropic Claude and Google Vertex are first-class alternatives. The customer’s security team picks the approved foundation model; the engine swaps without code changes.

How does the engine choose which model to run a task on?

The engine routes each task by complexity. It scores difficulty from signals like reasoning load, tool breadth, and conversation depth, then matches the task to a tier within your approved provider’s model family — frontier models for hard, high-stakes or audit-grade decisions, faster and cheaper models for routine cases. Routing is deterministic and explainable: every decision records why a tier was chosen, in the same audit chain as the decision itself. Compliance-sensitive workflows can pin a minimum tier so even easy cases run on a capable model.

Is Vihaya SOC 2 or HIPAA certified?

Vihaya is pre-certification. Audit envelopes self-map to SOC 2, HIPAA, and ISO controls so evidence collection is a side effect of running. Customers retain certification ownership for the deployed system inside their tenant.

What data does Vihaya see during a pilot?

Vihaya engineers see only sandbox or de-identified data unless the customer’s data-fiduciary controls grant otherwise. Production data never leaves the customer tenant. BAA/DPA are signed in week 1 before any access is provisioned.

What does a Vihaya decision look like?

Each Vihaya decision is a typed, append-only record: outcome, confidence score, rationale, citations pointing to exact source passages, model, prompt hash, timestamp, reviewer if escalated. Reviewable from cold storage years later.

How is Vihaya different from SaaS AI vendors?

SaaS AI vendors run the model on their infrastructure; the customer sends data out. Vihaya deploys into the customer’s tenant, runs in their region, and emits audit-chained decisions the customer fully owns. No SaaS egress, no shared-tenant risk.

Can a customer leave Vihaya after the pilot?

Yes. The deployed service runs in the customer’s tenant on the customer’s contract with the model provider. Source artifacts, eval datasets, and audit envelopes belong to the customer. Vihaya hands over a runbook at week 12.

Does Vihaya do loan underwriting and AML review?

Yes. Vihaya supports loan underwriting and AML alert review for banks and NBFCs. The engine reads the borrower file, policy bulletin, and prior decisions, returning one cited, escalation-protected output per case.

How do I start a Vihaya engagement?

Book a 30-minute discovery call at vihaya.ai/contact. Vihaya identifies the highest-pain wedge in the workflow, maps the first decision contract, and sends a draft statement of work within a week. Design-partner slots are open as of May 2026.

Forward-deployed AI for regulated enterprises

Vihaya is a forward-deployed AI decisioning firm for regulated enterprises. The Vihaya Engine runs inside the customer’s own cloud VPC and emits immutable, typed decisions with citations to source passages. Cases below a configured confidence floor escalate to a human reviewer. Vihaya is pre-revenue; design-partner pilots are open.

A forward-deployed AI firm with a compounding engine underneath. We embed with one enterprise per engagement and ship a deployed AI service into their tenant — every decision cited, escalation-protected, audit-chained.

12 wks: Fixed-price pilot
6: Engine substrates
1: Tenant, your region
0: SaaS egress

The output shape

Not a chatbot reply — a typed, cited decision record

Every Vihaya engagement emits decisions in this shape: an outcome, a confidence score, a rationale, and citations pointing to the exact source passages the agent grounded on. Append-only. Reviewable years later from cold storage.

outcome: Machine-readable verdict — not prose
confidence: Numeric. Below threshold → routes to your reviewer
rationale: Why this outcome, in your domain language
citations: Exact passages — policy clauses, guidelines, prior cases
audit: Inputs, model, prompt hash, timestamp, reviewer if any

decision · case_id 84c2a1v1 · append-only

Outcome

Approveauto-adjudicated

Confidence0.94

0.0↑ threshold 0.851.0

Rationale

Claim falls within policy §4.2 covered-conditions list. Pre-authorisation matches treatment code. No exclusions trigger. Prior adjudications for similar codes: 12 of 12 approved within the last 90 days.

Citations

policy.v2024 §4.2preauth.94c1 line 7guideline.IRDAI-circ-09 §3history.codeT45 (12/12)

model: gpt-4o · prompt#a91f · run 2.41sreviewer not required

Representative output shape · field names and values change per engagement

The compounding engine

A request becomes a cited decision

In-tenant

Multi-cloud

Any LLM

Hybrid Retrieval

Agent Memory

Inline Citations

Live Guardrails

Typed Decisioning

Vihaya

Agent

Docs

Databases

APIs

Vectors

OpenAI

Claude

Gemini

Llama

Local

Durable agents on a context-mesh engine, running on your infrastructure. Every request climbs the stack and leaves as an audited, cited decision.

01 · The Engine

One substrate — any sector

Every engagement we ship talks to the same engine. Six substrates deployed inside the customer’s tenant. The customer keeps their data; the engine compounds across engagements.

Context Mesh

Typed knowledge graph + hybrid retrieval + episodic memory. Every entity, relation, and prior decision queryable in one place.

Decisioning

Immutable, cited decisions. Every output links back to the source passages the agent grounded on.

Durable Agents

Stateful, resumable agent runs with checkpoints. Long-running workflows survive process restarts.

Compliance

Audit envelopes self-mapped to SOC 2 / HIPAA / ISO controls. Evidence collection is a side effect of operation.

Adaptive Guardrails

Policy enforcement that learns from reviewer override signal. Tightens in the direction your experts disagree.

Agent Sandbox

Isolated execution for untrusted code and tool calls. The agent runs generated code without compromise.

02 · Early roadmap

Verticals we’re building toward

Design-partner pilots open on these wedges. Each engagement is 12 weeks, deployed into the customer’s tenant.

Claim adjudicationTelecom CXLoan underwritingAudit-safe deployment

01Health Insurance · BFSI

Design partner

Claim adjudication, automated

Agentic decisioning grounded against policy wordings + regulatory guidelines. Citation-backed, audit-chain protected.

Open

02Telecom

Design partner

Unified customer experience

Intent-driven CX automation at large-operator scale. Replace fragmented IVR + bot stacks with one agentic platform.

Open

03BFSI · Banks & NBFCs

Design partner

Loan underwriting & AML review

Reads the borrower file, policy bulletin, and prior decisions. One cited, escalation-protected output.

Open

Selected patterns, not a full catalog.

Each one ships into the customer’s tenant.

03 · What buyers evaluate

What your reviewers, compliance team, and auditors would see

Behaviour of the platform — not the implementation underneath.

Cited decisionsEscalation floorCompliance-alignedDeploys in your tenant

Cited decisions

Every decision references the exact policy clauses, clinical guidelines, or transaction patterns the agent grounded on.

Confidence-floor escalation

Below the configured threshold, cases route to your existing reviewers with our recommendation attached.

Compliance-aligned by design

Append-only audit, configurable PII redaction, data-fiduciary controls baked in.

Quality gates that block deploys

CI-blocking regression tests on hand-curated golden datasets. Drift → the deploy never reaches production.

Cost-aware model routing

The engine scores each task and matches it to the right model — frontier models for hard, high-stakes decisions, faster and cheaper ones for routine cases. Same audit chain, every tier.

Model-agnostic, multilingual

Use the foundation-model provider your security team has approved. Multilingual at the languages your customers speak.

Deployed in your tenant

Runs in your cloud, in your region. No SaaS egress. We embed; you run.

04 · Method

12 weeks — we embed, the system stays yours

Not a SaaS. A deployed service in your environment, with your team in the loop from week one.

Discovery & accessIngest & integrateShadow runHandoff

Weeks 1–2

Discovery & access

BAA / DPA signed. Sandbox credentials. Policy corpus delivered. SME interviews.

Weeks 3–8

Ingest & integrate

Service deployed into your environment. Policies indexed. Eval dataset built with your SME.

Weeks 9–11

Shadow run

Runs against shadow traffic. Weekly decision review. Threshold locked to your tolerance.

Week 12

Handoff

Runbook. Reviewer training. Success-metrics report. Go / no-go on production.

05 · Frequently asked

Straight answers

Last updated May 28, 2026

Vihaya is a forward-deployed AI systems firm. We embed with one regulated enterprise per engagement and ship a deployed AI service into the customer’s tenant, built on the Vihaya Engine — six substrates: Context Mesh, Decisioning, Durable Agents, Compliance, Adaptive Guardrails, Agent Sandbox.

Next step

The fastest way to see whether this fits your book is to run a pilot

No procurement maze, no platform license. One workflow, one tenant, one fixed-price engagement — so you can judge the output before you commit.

Read the PA case study

How it starts

01
Discovery call
30 minutes. We identify the highest-pain wedge and map the first workflow.
02
Draft SOW
A fixed-price, 12-week pilot — scoped to that one workflow. No open-ended retainer.
03
Deployed in your tenant
Your region, your cloud. Audit chain on every decision from day one.

Book a discovery call

30-minute call · draft SOW within days · no obligation