Engineering Lead (AI & Automation Products)

Dentsu Global Services; Bengaluru; On-site; INR 3,000,000 - 6,000,000; Full time

Job summary

Dentsu Global Services Bengaluru seeks a senior software architect to lead architecture decisions, own Postgres schema design, and manage Claude/API integration. You will define integration patterns, security controls, and auditability while guiding a India-based team.

You will drive full‑stack work with Python APIs and React UIs, set engineering standards, observability, and testing, and run sprint planning, code reviews, and mentorship to deliver reliable, reusable AI-enabled workflows.

Qualifications

• Proven experience leading software architecture in large portfolios.
• Strong Postgres data modeling and schema design.
• Experience with AI/ML workflow integrations and API design.

Responsibilities

• Lead architecture decisions across multiple streams and ensure alignment with product requirements.
• Mentor junior and mid-level developers based in India with code-quality focus.
• Define and enforce engineering standards, testing discipline, observability, and secure coding practices.

Skills

Architecture leadership

API design

Postgres schemas

RBAC/multi-tenant

Team leadership

DevOps / IaC

Tools

FastAPI

React

Claude Code

OAuth 2.0

Azure Key Vault

Observability tools

Job description

• Lead architecture decisions across the portfolio — schema design, API contracts, integration patterns, workflow orchestration, and when to integrate via direct API versus MCP (Model Context Protocol), weighing latency, control, security, auditability, and reusability tradeoffs
• Hold delivery accountability — scope, timeline, quality — across all active workstreams, in partnership with the Director who owns product requirements and prioritization
• Design and own the Postgres data schema underpinning the product (multi-table: pipeline data, access control, cost attribution, audit) — correctness and extensibility here is foundational, not incidental
• Build and own the Claude/API integration layer: system prompt design and testing, multimodal document ingestion for XLSX, DOCX, PPTX, and PDF via reliable extraction/conversion pipelines, structured output parsing, schema validation, and fallback handling
• Design practical automation workflows — background jobs, queues, retries, idempotency, human review points, run state, and replayable audit history for multi-step AI-assisted processes
• Build and own third-party DSP API integrations (e.g. DV360, TTD) — OAuth flows, structured data file formats, read/write scoping by phase
• Design and own secrets and credential management — API keys, OAuth credentials, Azure Key Vault or equivalent secrets manager; nothing in code or environment variables in source control
• Design and own multi-tenant access control — role-based permissions, client-scoped data visibility, self-service onboarding
• Design cost and usage attribution patterns — external API calls tagged and logged at the client and run level, feeding finance and program-level reporting as the portfolio scales
• Review technical output against product requirements — flag when implementation doesn't match the requirement or the architecture is wrong for the problem
• Manage a team of junior and mid-level developers based in India — hiring input, onboarding, performance, growth planning
• Run code reviews, pairing, and technical mentoring as a standing practice, not an occasional one
• Set and enforce engineering standards: testing discipline, data validation, AI evaluation discipline, observability, and code quality bar
• Standardize how the team uses Claude Code — establish shared conventions, prompt/context patterns, and reusable practices so the whole team improves its delivery quality, not just you
• Run sprint-level technical planning and unblock the team day to day; escalated cross-team blockers to the Director rather than letting them stall delivery

Build Execution

• Build full-stack applications end to end where needed: Python APIs (FastAPI or similar) and React/Tailwind front ends
• Use Claude Code as a daily driver — both for personal output and as the standard tool your team is held to
• Write tests, build data validation frameworks, and instrument observability across the pipeline, including prompt/model traces, structured-output validation, latency, cost, and failure analytics
• Build and maintain practical evaluation harnesses for AI behavior — golden test sets, prompt/model versioning, regression checks, failure taxonomies, and release gates for higher-risk automations