Principal AI-Native Full-Stack Architect

Experience:10+ years in software, including 3+ as an architect or tech lead on multi-tenant SaaS, and atleast 6+ months shipping production software mainly through AI coding agents.

The role.You design the Spektra’s SaaS modular platforms and the AI-driven engineering system that builds it. You own the boundaries between the .NET/Angular core and the Next.js/React/Python modules, the contracts between them, and the specs, templates and guardrails our coding agents work from. You lead by writing specs and reviewing output, not by typing code: every implementation is generated by agents and verified by you and your team.

• Platform architecture
  • Own the target architecture: core services (identity, tenancy, entitlements, lab engine, learner record) and independently deployed modules.
  • Define module contracts: manifests, APIs, events, shared header/shell, branding, and the per-module data boundary (own schema/database, no cross-module reads).
  • Make and record decisions as ADRs: edge routing (Azure Front Door), identity (B2C / Entra External ID, server-side sessions, per-module app registrations), data, and hosting.
  • Plan the safe migration off legacy V1 services onto vNext, one path or feature at a time, with rollback at every step.
• The agent-driven engineering system
  • Design how work flows from idea to production with agents: spec → plan → task breakdown → agent implementation → automated checks → human review → deploy.
  • Own the standing context agents load every session:AGENTS.md/CLAUDE.md, architecture rules, coding standards, skills, MCP servers and hooks. Review changes to them like code.
  • Maintain the golden module template (our TaskBoard reference module) so a new module goes from scaffold to QA in days.
  • Set the automated guardrails agents must pass: typecheck, lint, dependency-boundary rules, contract tests, row-level-security tests, secret scanning, conformance checks.
  • Decide which agent, model and mode fits which task, and track cost, speed and defect rates per workflow.
• Quality, security and operations
  • Review the highest-risk agent changes yourself: authentication, authorization, tenant isolation, payments/credits, infrastructure and data migrations.
  • Threat-model new modules and AI features (prompt injection, credential leakage, abuse of lab environments).
  • Define SLOs, observability and incident practice for the platform; make sure every environment can be rebuilt from infrastructure-as-code (Bicep/Terraform).
• AI product architecture
  • Architect AI-powered product features such as the AI Trainer (realtime voice, screen understanding, lab validation tools) and Lab Studio (AI-generated labs, guides and checks).
  • Choose and integrate model providers (Azure OpenAI, Anthropic and others), with evaluation suites, guardrails and cost controls.
• Leadership
  • Coach developers on spec writing, agent orchestration and critical review.
  • Work with product, the .NET core team and partners on roadmap and trade-offs; explain them clearly to non-engineers.

• Deep, current experience with both halves of our stack: ASP.NET Core/C#/SQL Server and Angular on one side; TypeScript, React/Next.js and Python on the other. You can read any of it fluently and spot what an agent got wrong.
• Proven architecture of multi-tenant SaaS: tenant isolation, RBAC/entitlements, white-labelling, API versioning.
• Strong identity and security fundamentals: OAuth 2.0/OIDC, Azure AD B2C or Entra, session design, OWASP Top 10, secrets management.
• Azure in production: Container Apps or AKS, App Service, Front Door or App Gateway, Key Vault, managed identities, Bicep or Terraform.
• Demonstrated daily use of AI coding agents (Claude Code, Codex, Cursor agent mode, Copilot agents or similar) to ship production systems, including multi-agent/parallel workflows.
• Experience writing specs and acceptance criteria precise enough that an agent can build from them, and standing instructions that keep a codebase consistent across many agent sessions.
• Clear written English: ADRs, specs and review comments are your main output.

• Built MCP servers, agent skills or custom agent tooling; experience with the Claude Agent SDK, OpenAI Agents SDK or similar.
• Built LLM products in production: RAG, tool use, realtime voice (WebRTC), evaluations.
• Learning-platform or lab-platform domain: LMS, LTI 1.3, assessments, cloud sandboxes, Guacamole/RDP consoles.
• Migrated a large legacy platform incrementally (strangler pattern).
• Microsoft, AWS or GCP architect certification.

Originally posted on Himalayas