Venkata Peetla

Dual track

FDE Field Method

One method for Principal Architect and Forward Deployed loops: clarify the workflow, ship a thin production wedge under real constraints, govern irreversible actions, then hand off ownership.

This is not a sixth mega-platform — it is the operating method that turns the existing 5-spine stack into customer-ready delivery proof.

Method

Discover → score → ship → govern → eval → handoff

  1. 01

    Discover

    Week one: jobs to be done, systems of record, SSO/ACLs, data reality, irreversible actions, change windows, success metric, exit owner. Skip this and you're building theater.

  2. 02

    Score the wedge

    Rank on value × feasibility × risk × reuse × named sponsor. Prefer ≤90-day measurable outcomes. Platform theater loses.

  3. 03

    Ship the walking skeleton

    Read/assist path first. Access-before-ranking retrieval. Integrate into their identity and event planes — not a forever demo VPC.

  4. 04

    Govern irreversible actions

    HITL and policy on writes that can't be undone. Decline-on-low-confidence beats fake autonomy every time.

  5. 05

    Eval before rollout

    Groundedness, tool success, HITL reject rate, and a baseline window before anyone claims $ savings.

  6. 06

    Handoff ownership

    Runbooks, prompt/eval owners, and a team that can operate after I leave the critical path. If that doesn't exist, the embed failed.

Week-1 questions

What I ask before proposing architecture

  • Who waits on what decision today — and what metric moves in 90 days?
  • Which system is source of truth for writes?
  • How does identity flow (SSO, groups, row/object ACLs) into retrieval?
  • Which actions are irreversible on day one?
  • What is the change-control window and who can approve production?
  • Who owns the system when I leave the critical path?

Embed lab

Acme Support Agent Embed

Named FDE wedge on the existing spine: IdP → SCIM → Strict RAG → Slack/Salesforce → HITL → tenant meters. Not a thirteenth SaaS product.

ADR-03220-min panel rubricCase study
  • OIDC + SAML panel logincode demonstrated
  • SCIM users/groups into AegisAIcode demonstrated
  • Strict ERAG + spoof-tenant break testcode demonstrated
  • HMAC webhooks + DLQ replaycode demonstrated
  • Slack recovery + Salesforce caseoperator wiring
  • Tenant health + FinOps freezecode demonstrated
  • Stripe test-mode usage previewoperator wiring
  • Evidence pack + post-mortem + commscode demonstrated

code_demonstrated = harness/API/tests. operator_wiring = live IdP/Slack/SFDC/Stripe/Strict twin still owner-configured (see operator wiring guide). Free-tier APIs may cold-start — PANEL_DAY_FREE_RUNBOOK.

Panel script
20 min
FDE moves mapped
12
Stripe
test mode only
RLS / HubSpot / GWS
refused / adapter-only
  • Refuse: No HubSpot/Google Workspace packs — adapter contract only
  • Refuse: No SOC2 attestation claims — compliance log pattern only
  • Refuse: No Postgres RLS theater — ADR-026 verified tenant + blast-radius

Proof layers

Same spine, FDE framing

AegisAI — gateway + HITL

Policy before side effects — the governance boundary I'd enforce on day one.

Enterprise RAG — access-aware

Access-before-ranking when customer data ACLs are non-negotiable.

VAP — orchestration

Multi-agent workflows as a deployable wedge — not a chat toy.

Honesty

Evidence rules panels can trust

  • Lucid Supply Chain / Commerce work is an internal customer embed — not invented third-party logos.
  • Open reference repos (AegisAI, VAP, Enterprise RAG, Content Factory) show how I build. They are not claims that Lucid runs those exact binaries.
  • Apple/Google work was client engagement via Sparity — not direct Google employment.
  • Acme Support Agent Embed (ADR-032) is a reference customer wedge on the spine — not a claim Acme is a real logo customer.