Job Description

About the AI Technology Innovation Centre

We are establishing a world class AI research center called “AI Technology Innovation Center” (TIC) in Bengaluru. The TIC will drive research in frontier AI technology with top AI talent and state-of-the-art infrastructure. The research breakthrough will lead to commercialization of next generation AI solutions. The AI TIC is being strategically designed to establish a globally recognized leader in AI research and development, driving innovation and setting industry benchmarks. The TIC aims to attract, nurture, and retain top-tier AI talent in India by fostering a dynamic, inclusive, and high-impact work environment that encourages continuous learning and breakthrough thinking.

By seamlessly integrating commercial agility with a pioneering scientific mindset, we will deliver scalable AI solutions that address real-world challenges while pushing the frontiers of technology. This ensures that our AI-TIC excels in cutting-edge research and translates discoveries into tangible business value, reinforcing our position as a hub for AI excellence.

The Role Mandate

As the Principal Engineer of AI TIC, you’ll own the end-to-end system architecture and design and delivery of production-grade agentic and Generative AI systems (on-premise/cloud). This is a highly hands-on role requiring deep architectural insight, coding proficiency, and an obsession with performance, scalability, and reliability. You’ll architect secure, cost-efficient AI platforms, guide developers through complex debugging and optimization, and ensure all systems are observable, governed, and production ready. You will create a reference platform and AI pipeline on which TIC research will get translated to real product.

͏

Why This Role - What You Gain

• Founding-stage platform ownership

Define architecture, standards and SDKs that every Centre team will build on.

• Direct link to global products

Carry frontier research into dependable, observable and secure deployed systems.

• The hardest form of the problem

Engineer memory, emotion control and self-learning under real compute and reliability constraints.

• Technical leadership without leaving code

Mentor, review, debug and ship as a true principal engineer.

The Larger Impact

• Determine whether empathic agents become dependable products rather than demonstrations.
• Create a secure, modular and cost-efficient platform for multimodal, memory-enabled, tool-using agents.
• Lead model adaptation, compression and optimisation for self-managed GPU environments.
• Create reusable SDKs, connectors, CI/CD templates, evaluation harnesses and observability.
• Set the engineering standards that become the Centre’s technical reputation.

͏

Key Responsibilities:

• Architect Production AI Systems: Design robust overall architectures for agentic systems (planning, reasoning, tool-calling), GenAI/RAG pipelines, and evaluation workflows. Create detailed design documents including flow/UML/sequence diagrams and AWS deployment topologies. Additionally, ensure architectures support advanced LLM training and inference workflows, incorporating distributed strategies for scalability.
• Optimize for Cost & Performance: Model throughput, latency, concurrency, autoscaling, CPU/GPU sizing, and vector index performance to ensure scalable, efficient deployments. Include optimization for multi-node GPU clusters and distributed training efficiency to reduce compute overhead.
• Lead Debugging & Stability Efforts: Conduct deep-dive debugging, fix critical defects, and resolve production incidents; pair-program with developers to improve code quality and performance. Apply MLOps-driven stability practices, leveraging configuration management and automated recovery for high availability.
• Standardize Agentic Frameworks: Build reference implementations using Semantic Kernel (preferred), LangGraph, AutoGen, or CrewAI with strong schema validation, grounding, and memory management.
• Implement Observability & Monitoring: Set up distributed tracing, metrics, and logging via OpenTelemetry and Datadog. Standardize dashboards, alerts, and incident response workflows.
• Govern Evaluation & Rollouts: Build test and evaluation frameworks—golden sets, A/B experiments, regression suites, and controlled rollouts—to ensure consistent quality across releases.
• Establish Engineering Standards: Create reusable SDKs, connectors, CI/CD templates, and architecture review checklists to promote consistency across teams.
• Cross-Functional Leadership: Collaborate with product, data, and SRE teams for capacity planning, DR strategies, and post-incident RCA reviews. Mentor engineers to strengthen design and reliability practices

How the Centre Works

The TIC follows a research-to-product operating model. Scientists and engineers work as one team across hypothesis formation, algorithm design, model development, evaluation, platform engineering and technology transfer. Every role combines technical depth with ownership, reproducibility, responsible AI and clear communication of outcomes.

͏

Qualifications and Experience

Experience and leadership

• Education: Bachelor’s/Master’s from a top-tier institute (IIT/Tier-1) in Computer Science, AI, or related field.
• 7–10 years in software/AI engineering, including 4+ years in GenAI application development and 2+ years architecting agentic AI systems.
• Expert in Python 3.11+ (asyncio, typing, packaging, profiling, pytest).
• Hands-on experience with Semantic Kernel, LangGraph, AutoGen, or CrewAI or equivalent.
• Proven delivery of GenAI/RAG systems on AWS Bedrock or equivalent vector-based platforms (OpenSearch Serverless, Pinecone, Redis).
• Deep understanding of AWS ecosystem: EKS, Bedrock, S3, SQS/SNS, RDS, ElastiCache, Secrets Manager, IAM/Okta, Kong API Gateway management and automated recovery for high availability.

What Success Looks Like (18-24 Months)

• A hardened, observable agentic system and platform having custom models on on-premise infrastructure.
• Reference implementations, SDKs and CI/CD templates become the Centre default.
• Quality, safety, latency and cost are measurable and continuously improved.
• Research lands in production safely through a regular, governed release cadence.
• Production problems are detected early and converted into systemic improvements.

͏