SEPANTA / Finspector.AIcloud & devopsbackendagentic AI

Finspector AI-Driven Financial-Marketing Compliance Platform

Helping financial firms detect and manage regulatory risk in their marketing, with an agentic pipeline whose own decisions must be defensible.

My roleCloud & DevOps Engineer, with backend contributions

Team~6 engineers

MarketB2B, financial firms in the UK (US a future market)

RuntimeCloud Run · Pub/Sub · Cloud SQL · Valkey · Google ADK

01Overview

Finspector is a B2B platform that helps financial firms detect and manage regulatory risk in their marketing. It automatically inspects financial promotions across files, social-media posts and channels, and websites against regulatory requirements, whether FCA standards or custom rules tailored to each organisation, then flags compliance issues and maintains an audit trail.

Architecturally it is a microservice pipeline coordinated by an agentic "Brain": specialised agents (file processor, social-media processor, web scraper) collect and pre-process content, while the Brain routes work between them and performs the AI analysis. Services run on Cloud Run, communicate asynchronously over Pub/Sub, and persist to Cloud SQL. The agents are built with Google's Agent Development Kit (ADK).

Figure 1. Agentic microservice architecture: source agents feed a coordinating Brain that produces compliance findings and an audit trail.

02My contributions

Cloud & DevOps primary role

Keyless CI/CD: built the GitHub Actions delivery pipelines authenticating via Workload Identity Federation (no service-account keys), covering build and deployment.
Automated security gating: dependency / vulnerability checks and Docker image vulnerability scanning on the path to deploy.
Public edge & networking: designed the load-balancer flow and domain management, and added Cloud Armor protection at the edge.
Access & identity: IAM management across the project.
Observability: deployed Langfuse to GKE via Terraform for LLM tracing, plus logging, tracing and alerting.

Backend contributions

Worked as a developer on the social-media processor service, one of the agentic pipeline's source processors.
Built the dependency layer connecting services to the Valkey cache.

Figure 2. The secure delivery pipeline and runtime edge I built, with end-to-end observability.

03Technology

Python · FastAPIGoogle ADK (agents)Cloud RunPub/SubCloud SQLValkeyGitHub ActionsWorkload Identity FederationCloud ArmorCloud Load BalancingTerraform · GKELangfuse

04Impact

Gave an AI compliance product keyless, secure, automated delivery with vulnerability gating before every deploy.
Hardened the public edge with Cloud Armor and managed load balancing, and centralised IAM.
Made the agentic system observable and auditable (Langfuse LLM tracing, logging, tracing, alerting), essential for a tool whose own decisions must be defensible.
Contributed to the social-media processor service and built the shared Valkey cache integration.

Architecture described at a conceptual level; internal identifiers and credentials are intentionally omitted.