Streaming 24h · Cash session shut · Connecting · --:--:-- UTC

USD/INR—0.00%EUR/INR—0.00%GBP/INR—0.00%JPY/INR—0.00%AUD/INR—0.00%SGD/INR—0.00%CHF/INR—0.00%HKD/INR—0.00%

Architecture

CorpVidesh AI
Legal architecture.

Modular decision services connect Indian law, market evidence and secure filings. Every interpretation identifies its statutory basis, evidence gap and next accountable action.

Presentation library / revised 31 October 2026

Legal architecture & investor presentations.

Latest available PDFs, with dated editions and earlier dossiers retained.

Investor brief

01 / 1 page · A3 · 2026-10-31

Synopsis with Infographics

200-word abstract, connected workflow infographic, coverage boundaries and the five-member team.

PDF

02 / 2 pages · 16:9 · 2026-10-31

System Architecture & Functionality

Decision stack, identity boundaries, secure records, five-stage flow and production integration gates.

PDF

03 / 22 pages · 16:9 · 2026-10-31

Venture Capital Pitch Deck

Connected Console, 3D clearance, regulator/GIFT/fund portals, AI, global markets, security, evidence and pilot economics.

PDF

The three 31 October 2026 editions explain the latest implemented prototype features, workflows and technology, with clear simulation and production boundaries. PDF dates describe the edition, not verified current law or a fresh full-platform audit. Earlier editions and dossiers remain archived and may describe superseded features. Official filing access, regulator endorsement, certified security and granted patents are not implied.

Full presentation desk

Proof of feasibility / evidence & delivery

Evidence before scale.

Retrospective proof-of-concept acceptance baseline, recorded 26 October 2026. Measured technical checks are separated from blocked pilot criteria. The overall production gate remains open.

01 / Executable proof

Test the safeguards.

Pure policy checks run now. Saved regression evidence retains its original measurement date. Neither certifies production readiness.

02 / Pipeline → outcome

LangChain building blocks + LangGraph control

Open working desk
  1. 01

    Cited corpus

  2. 02

    Passages

  3. 03

    Embed / lexical fallback

  4. 04

    Rank

  5. 05

    Grade

  6. 06

    Draft + QA

Outcome / Cited draft or explicit human-review outcome

The current statutory corpus is bundled and indexed in-process, not a persistent vector database or a general uploaded-document ingestion service. Low-quality retrieval branches into query rewriting and local expansion.

03 / Stage gates

Available

Prototype

Does the idea feel possible?

Interactive desks, input scenarios, 3D processing simulations and workflow comparisons.

Demonstrations must remain labelled, not presented as completed transfers.

Partially evidenced

Proof of concept

Does it work on real inputs?

Provider FX observations, cited retrieval, service regression tests, quality gates and controlled cost planning.

Still open: signed-in save/read-back, multi-account isolation and representative live AI evaluations.

Not approved

Production

Can it run reliably for users?

Security, monitoring and CI foundations exist inside a single deployment.

Still open: load SLOs, durable checkpoints, recovery drills, pilot feedback, independent assurance and authorized external integrations.

04 / Model usage sensitivity

Editable planning inputs, not a provider quote. Rates start at zero until supplied from the workspace model catalog. Excludes embeddings, tool calls, storage, licensing and taxes.

Projected model usage: Awaiting rate assumptions

Decoupled by contract

Four bounded service domains; validated request contracts, separate policy logic and server-only execution.

Law before automation

Investor eligibility, instrument classification, INR distinctions and statutory evidence precede route selection.

Authority stays explicit

Company-scoped filings and private agent outcomes. Official regulator approval and bank execution remain unconnected.

01 / Service topology

Treasury / regulator deskValidated server functionsBounded domain servicesCloud evidence / providers

01 / Legal interpretation

legal-route.v1

Validated facts → cited checks → human review

Running:
Server assessment + statutory graph
Scale design:
Stateless evaluation; versioned legal releases
CI gate:
Independent deterministic test job; no paid model calls.

02 / Corporate filings

form-fc.v1

Verified account → company membership → append-only record

Running:
Authenticated server functions + Cloud RLS
Scale design:
Indexed company reads; managed storage
CI gate:
Independent deterministic test job; no paid model calls.

03 / Market evidence

market-evidence.v1

Provider adapter → timestamp → provenance → unavailable state

Running:
Shared FX / market / observed-history functions
Scale design:
Shared provider cache; no fabricated evidence
CI gate:
Independent deterministic test job; no paid model calls.

04 / Agent orchestration

agent-outcome.v1

Verified viewer → bounded memory → QA → private outcome

Running:
App-hosted graph + read-only tools
Scale design:
Request-scoped execution; durable private outcomes
CI gate:
Independent deterministic test job; no paid model calls.

Deployment reality: these domains run within one application deployment. Independent service CI is configured; separate microservice hosting, queues, autoscaling guarantees and load-test certification are not provisioned. Splitting deployment requires infrastructure and service-to-service identity, not just a diagram.

02 / Statutory route assessment

A route is a legal conclusion.
Not a switch.

Non-persisted advisory triage under the 2022 overseas investment framework. No company identity or confidential documents are collected here. Current amendments, complete permissions and actual filing dates require counsel / AD-bank verification.

Interpretation contract / legal-route.v1

Facts → rule → evidence → next action

  1. 01Classify ODI/OPI, investor, activity and control.
  2. 02Aggregate existing exposure and assess the applicable reference ceiling.
  3. 03Identify prohibitions, approvals, NOC and evidence holds.
  4. 04Separate tax, corporate authority, AML and IFSC obligations.
  5. 05Return cited findings, a trace ID and an accountable next action.

03 / Indian statutory source register

FEMA is the 1999 Act; the overseas investment Rules, Regulations and Directions were introduced in 2022. These are source references and bounded summaries, not a verified consolidation of every amendment as of today. Form FC, APR, OPI and FLA obligations depend on the facts; not every OPI transaction uses Form FC.

FEMA, 1999

Capital-account framework; sections 6, 10 and 13

Permission, authorised-person duties and enforcement; not an ODI Act amended in 2022.

Overseas Investment Rules, 2022

Definitions, Rule 19, Schedules I and II

Investor and instrument classification, financial commitment, OPI eligibility and prohibited activities.

Companies Act, 2013

Sections 179 and 186; applicable exemptions

Board authority, corporate investment/loan/guarantee approvals and disclosures; FEMA limits are not a substitute.

PMLA and RBI KYC directions

Customer due diligence; reporting-entity obligations

Beneficial owner, source-of-funds and sanctions checks. Do not label every corporate transfer an FIU approval.

IFSCA regulatory framework

IFSC-specific regulated activities and OI Schedule V

Apply only where an IFSC entity/activity is involved; no blanket six-regulator approval mandate.

04 / Differentiation with evidence

Indian law. Live market context.
One accountable decision record.

The differentiated proof of concept combines legal retrieval, conditional treasury scenarios, company-scoped filings and supervisor evidence. “First of its kind” is a positioning thesis, not a verified priority, patent, market exclusivity or regulatory endorsement claim. Six regulator lanes are a demonstration model, not six mandatory approving authorities.

Connected execution · private evidence

One handoff. One accountable record.

AI functions

Checking…

Private run history

Public demonstration · no anonymous inputs stored

Licensed FX

Public provider feeds · licensed access pending

External approvals

Not connected · filings and regulator lanes remain simulations

Verified users receive owner-scoped, append-only agent outcomes; recent successful summaries ground subsequent copilot and statutory requests. Public demonstrations remain unsaved. Local demonstration roles never authorize private records or external actions. Managed graph deployment and official submissions require separate provisioned connections.

01 · Context

Context engineering

Facts, recent determinations and the dialogue tail are compressed into a bounded preamble instead of a growing transcript, so cost per turn stays flat as a thread ages.

02 · Retrieval

Vector RAG with a corrective loop

The graph and this probe share semantic retrieval with lexical fallback. A grader scores passages, rewrites the query and expands the local corpus before drafting; no external statutory lookup is claimed.

03 · Reasoning

Multi-agent graph

A coordinator routes to retrieval, grading, rewriting, drafting and quality checks. Streamed model chains run inside the application; node-resume checkpoints are not provisioned.

04 · Memory

Three-tier agent memory

Device-local turns and facts accompany the graph; verified users also receive private server-side run outcomes and bounded context shared across agents.

05 · Tools

Scoped tool boundary

FX, markets and statutory searches execute read-only server functions. Private records require verified identity; unconnected mutations and official submissions are refused.

06 · Evaluation

Regression gate

The suite asserts on the structured filing packet — route, forms, citations, quality gate, latency budget — so a behavioural regression fails a case, not a vibe check.

07 · Operations

Observability and governance

Span-level latency and usage accompany graph outcomes. Verified users retain private outcomes; tracing requires credentials and regulators do not receive official access through a demo role.

08 · Surface

Prediction, not reporting

Every desk opens with a forecast and the action that improves it. Open the prediction surface.

Retrieval — semantic, with a deterministic fallback

The corpus is embedded once per process and searched by meaning; if the embeddings endpoint is unreachable the same query is answered lexically rather than left ungrounded.

Agent memory — 0 facts · 0 runs · 0 turns

Semantic facts

Set an entity and a corridor on the prediction surface — the desk remembers them.

Episodic runs

  • Runs from the agent graph land here and are replayed into the next question.

Compiled context sent to the model

empty — nothing to carry forward yet

Server-connected tools · demo role “guest” grants no private access

ToolServer / transportScopeAuthEffectEgress
fx.ratecorpvidesh.fx · httpfx.readnone (public feed)readNone — currency codes only leave the rail.
market.quotecorpvidesh.markets · httpmarket.readnone (public feed)readNone — index symbols only.
statute.searchcorpvidesh.corpus · in-processcorpus.readsession bearerreadNone — the corpus never leaves the process.
ledger.readcorpvidesh.ledger · in-processledger.readsession bearerreadCase metadata stays inside the tenant boundary.
ledger.stampcorpvidesh.ledger · mcp/httpledger.writeservice credentialwriteHash-chained entry; payload hashed, not copied.
filing.submitcorpvidesh.filing · mcp/httpfiling.submitofficer step-upirreversibleForm 15CA/15CB payload to the statutory endpoint.

Read tools execute on the server. Private history requires verified identity; unconnected external actions are refused. This activity list is device-local, not an official audit ledger.

Evaluation and managed deployment

Managed deployment — LangGraph Platform

graph “statutory” · langgraph.json

Runtime

Application-hosted LangGraph via server functions. Managed LangGraph deployment is not connected.

Scaling

Request-driven application runtime; no managed graph queue or background resume is provisioned.

Persistence

Verified users: private append-only run outcomes in Lovable Cloud. Node-resume checkpoints are not configured.

Observability

Per-run spans in the desk and private run history; LangSmith mirroring requires a configured credential.

Assistant configurations · managed deployment not connected

  • Statutory determination

    asst_statutory_full

    Full run: coordinate, retrieve, grade, draft the filing packet, quality-check every citation.

    recursion_limit 24 · streaming on · corrective retries 2

  • Corrective retrieval probe

    asst_retrieval_only

    Retriever + grader + rewriter only — used to test the corpus without spending draft tokens.

    interrupt_after ['grader'] · streaming on

  • Filing QA gate

    asst_qa_gate

    Re-runs the quality agent against an existing draft before a filing leaves the desk.

    entry ['qa'] · deterministic temperature 0

The factory at src/lib/agents/platform-graph.server.ts:makeDeployedGraph is the same builder the desk runs in-process, so a managed run and a desk run execute an identical topology.

Graph regression suite — 4 cases against the application graph

Not run in this session

Qwen · reasoning

Statutory reasoning model reads the rule and the filing together, and writes the conclusion with the rule quoted.

LangChain · orchestration

Chains the retrieval, tool calls and checks in a fixed order so every filing is examined the same way.

Alibaba Cloud · compute

Runs the heavy document and screening workloads in an India-resident region.

Federated ledger

Each result is hash-chained into a block every regulator node holds a copy of.

Human desk

A named officer signs this off. The machine only prepares the file.