Veni AI

Veni AI Sustainability Cockpit: Open-Source ERP-to-Report Sustainability Reporting

Veni AI Sustainability Cockpit combines ERP connector sync, canonical fact modeling, retrieval, verification, and controlled publish into one open-source sustainability reporting workflow.

The product is designed for teams that need more than a static ESG dashboard. It turns connector-fed data into audit-ready report packages with evidence, calculation traces, and tracked artifacts.

The architecture keeps SAP, Logo Tiger, and Netsis integrations separate at the connector layer, then normalizes every record into a canonical fact shape before narrative generation or packaging begins.

Fail-closed trust rules ensure the system does not publish unsupported claims: no evidence, no claim; no calculator artifact, no numeric claim; no verifier pass, no publish.

Sustainability Cockpit
Open Source Sustainability Reporting

ERP-to-package disclosure workflow

Turn ERP data into auditable sustainability reports

Built for teams that need SAP, Logo Tiger, and Netsis signals to become publish-ready TSRS and CSRD report packages.

The system synchronizes ERP inputs, normalizes every accepted record into canonical facts, verifies claims against evidence and calculation traces, and only publishes when package gates stay green.

Open SourceSAP / Logo / NetsisTSRS / CSRDControlled publish
Connector coverage
3 ERP inputs
Package pipeline
9 stages
Tracked outputs
6 artifacts
Trust policy
Fail-closed

Connector Matrix

Certified ERP inputs, one canonical fact model

Each connector preserves its own delta semantics and ingest contract, but every accepted record is normalized into a traceable disclosure object before it can influence narratives or package status.

01delta_token

SAP / OData

OData pull with ERP-grade freshness tracking for energy, emissions, and governance signals.

Best suited for climate, electricity, and governance metrics that need source-system lineage and repeatable sync jobs.

  • E_SCOPE2_TCO2E
  • RENEWABLE_ELECTRICITY_SHARE
  • BOARD_OVERSIGHT_COVERAGE
02snapshot_watermark

Logo Tiger / SQL View

Read-only SQL View sync for workforce, safety, and committee cadence metrics.

Captures operational and people metrics without writing back into the ERP, while keeping replay and support diagnostics visible.

  • WORKFORCE_HEADCOUNT
  • LTIFR
  • SUSTAINABILITY_COMMITTEE_MEETINGS
03cursor_or_updated_at

Netsis / REST

REST pull for supplier, materiality, and stakeholder engagement disclosures.

Supports supply chain and disclosure-planning flows where cursor-based deltas and API-level diagnostics matter.

  • SUPPLIER_COVERAGE
  • MATERIAL_TOPIC_COUNT
  • STAKEHOLDER_ENGAGEMENT_TOUCHPOINTS

Canonical + Trust

Every disclosure starts with a traceable data object

Records do not flow straight into narrative text. They become canonical facts that carry business meaning, ownership, unit discipline, and evidence-ready lineage before any disclosure logic runs.

Canonical fact object
{
  "metric_code": "E_SCOPE2_TCO2E",
  "period_key": "2025",
  "unit": "tCO2e",
  "value_numeric": 12450,
  "source_system": "sap_odata",
  "source_record_id": "sap-scope2-2025",
  "owner": "[email protected]",
  "confidence_score": 0.98,
  "trace_ref": "sap://scope2/2025"
}

What every fact carries

Period-aware facts

Every metric is anchored to a reporting period so year-on-year analysis and disclosure scope stay explicit.

Unit normalization

Units are standardized before package composition so emissions, ratios, counts, and rates stay comparable.

Ownership and confidence

Each fact records who owns the metric and how confident the system is in the normalized output.

Traceable disclosures

Source ids and trace refs keep each claim connected to the originating ERP object or evidence lane.

Fail-closed publish rules

01

No evidence, no claim

Narrative statements must resolve to evidence-backed citations before they become package-eligible.

02

No calculator artifact, no numeric claim

Numeric language is blocked unless the system can point to a deterministic calculation artifact.

03

No verifier pass, no publish

Critical FAIL states and missing approval-chain conditions stop controlled publish until the blockers are cleared.

Package Flow

A fail-closed flow from sync to controlled publish

The system treats publish as a tracked package pipeline rather than a blind PDF shortcut. That means stage history, artifact completeness, and verifier readiness remain visible all the way to release.

  1. 01

    sync

    Pull certified ERP connector data into project-specific run context.

  2. 02

    normalize

    Transform source rows into canonical facts with units, owners, and trace refs.

  3. 03

    outline

    Map facts and evidence into disclosure-ready section scaffolds.

  4. 04

    write

    Generate narrative drafts only after structure and context are in place.

  5. 05

    verify

    Run claim-level citation and numeric checks before package completion can continue.

  6. 06

    charts_images

    Prepare decorative visuals, charts, and tracked visual manifest slots.

  7. 07

    compose

    Assemble the PDF and supporting JSON exports from verified package state.

  8. 08

    package

    Register output metadata, stage history, and artifact inventory.

  9. 09

    controlled_publish

    Release only when verification, approval, and artifact gates all stay green.

Disclosure frameworks in scope

TSRS1

Governance and risk-management disclosure flows align the board, policy, and oversight story.

TSRS2

Climate and energy sections focus on emissions, electricity mix, and operational reduction narratives.

CSRD

Workforce and supply-chain narratives connect social, materiality, and supplier coverage metrics.

Why the flow matters

Publish model

Tracked package jobs, not instant blind downloads

Trust posture

Evidence-backed, calculation-aware, approval-gated

Operator view

Status, lanes, artifacts, and verifier pressure stay visible

Product Surfaces

Five operator surfaces for evidence, review, and release

The open-source product already exposes the reporting workflow as a working control room: from new run setup to retrieval diagnostics, evidence ingest, and package release.

01

Executive Report Factory Dashboard

Connector freshness, KPI strip, verifier pressure, package lanes, artifact health, and recent run movement in a dense executive cockpit.

The open-source product already exposes the reporting workflow as a working control room: from new run setup to retrieval diagnostics, evidence ingest, and package release.

02

New Report Run Wizard

A report factory launchpad for company profile, brand kit, framework scope, governance owners, and connector selection.

03

Evidence Ingest Workbench

Upload source files, run OCR or queued extraction, inspect index health, and keep raw payload visibility for evidence operations.

04

Retrieval Research Bench

Tune hybrid retrieval queries, scoring thresholds, period hints, and disclosure tags before claims move toward publish.

05

Controlled Publish Board

Execute runs, clear triage pressure, refresh package status, and download tracked artifacts from one release surface.

Report Output + OSS

Generated report output plus an open-source delivery spine

This is not just a dashboard concept. The repository already contains generated report previews, package artifacts, and a multi-service implementation spanning web, API, worker, data, and AI services.

Generated cover preview

Generated governance page

License

MIT

Monorepo

Web + API + Worker + Connector Agent

Local runtime

Docker Compose ready

Tracked package artifacts

report_pdf

Final PDF package with report narrative, bookmarks, and polished document output.

visual_manifest

A tracked inventory of decorative visuals, source type, and AI-generation flags.

citation_index

Claim-to-evidence mapping export for traceability and audit review.

calculation_appendix

Numeric calculation references used to justify quantitative statements.

coverage_matrix

Section coverage summary across required metrics and appendix references.

assumption_register

Structured list of assumptions captured during package generation.

Implementation stack

Frontend

  • Next.js App Router
  • React 19 + TypeScript
  • Tailwind CSS
  • ECharts and premium dashboard patterns

Backend

  • FastAPI
  • Pydantic
  • SQLAlchemy
  • Alembic
  • ARQ worker

Infra + AI

  • PostgreSQL
  • Redis
  • Docker Compose
  • Azure services
  • Connector agent for customer networks

Open-source by default

The public repository exposes connector onboarding, canonical fact sync, report generation, retrieval, verification, packaging, and downloadable artifacts. It is positioned as a real implementation baseline rather than a marketing mockup.