AI Architect

Globespan Solutions; Hyderabad; On-site; INR 4,200,000 - 7,000,000; Full time

Job summary

Globespan Solutions is seeking an AI Architect to design and build the core intelligence engine and conversational agents for a highly grounded finance platform. You will set technical direction for the engine and agent layer, collaborating with US-based AI leadership.

The role is hands-on and requires leading architecture decisions, ensuring reproducibility, guardrails, and verifiable data sources. Expect multi-cloud deployment and tight collaboration with platform engineering.

Qualifications

• Senior or staff-level applied AI/ML experience shipping production systems.
• Deep fluency in modern LLM patterns: RAG, tool-use/agents and structured output.
• Strong systems architecture skills for engine, data interfaces and agent layer.
• Demonstrated rigor in grounding, hallucination control and evaluation for LLM/agentic systems.
• Experience designing for reproducibility and determinism on probabilistic models.
• Strong production Python engineering with robust, tested services.
• Understanding why determinism and grounding matter in high-stakes domains.

Responsibilities

• Architect and build the intelligence engine that turns data into grounded, deterministic answers.
• Design conversational AI agents that serve finance users and orchestrate calls to the engine.
• Design and implement grounding, guardrails and deterministic pathways with source-backed results.
• Develop multi-cloud, in-VPC deployment supporting AWS, Azure and GCP environments.
• Provide technical leadership, review architecture and mentor engineers.

Skills

Senior AI/ML experience

LLM patterns

Systems architecture

Grounding & determinism

Python production services

Stakeholder communication

Agent orchestration

Tools

Bedrock

Vertex AI

Azure OpenAI

Python tooling

Job description

Job Summary

We are looking for an AI Architect to design and build the intelligence engine at the core of the platform, along with the conversational AI agents that sit on top of it. The engine produces answers that are computed, reproducible and traceable to verified source data, rather than improvised by a model. The agent layer interprets user questions, routes them to the engine and presents results in natural language without inventing the substance of the answer.

This is a hands‑on architecture role. You will set technical direction for the engine and agent layer and work closely with US‑based AI leadership.

Key Responsibilities

Intelligence Engine Architecture

• Architect and build the core engine that turns connected financial data into grounded, deterministic answers.
• Design the structured knowledge layer: how facts, relationships and context are associated and queried.
• Build deterministic reasoning paths with a clear chain back to source data.
• Define interfaces between the engine and the data, model and conversational layers.
• Author architecture decision records (ADRs) for the AI layer.

Conversational & Agentic Layer

• Design conversational AI agents that serve answers to finance users.
• Build agentic orchestration (e.g., LangGraph) that interprets intent, calls the engine and tools, and assembles responses, with verification and human‑confirmation checkpoints.
• Keep the agent layer a thin, auditable interface that routes and narrates while the engine supplies the substance.
• Design fallbacks and targeted clarifying questions for cases where context is missing.

Grounding, Determinism & Guardrails

• Make grounding a hard gate, with no fabricated or unsupported output.
• Use structured output, retrieval grounding and guardrails so generated text states only what verified data supports.
• Engineer for reproducibility in answer generation despite probabilistic models.
• Work with the evaluation team to gate faithfulness, hallucination and grounding in CI.

Multi‑Cloud & In‑VPC Deployment

• Design the AI layer to run inside the customer’s environment across AWS, Azure and GCP.
• Integrate managed LLMs in‑VPC (Amazon Bedrock, Google Vertex AI, Azure OpenAI) so customer data never leaves its boundary.
• Partner with Platform Engineering on model serving, scaling and cost‑aware inference.
• Account for single‑tenant and on‑premises constraints in architecture decisions.

Technical Leadership

• Set direction and standards for the engine and agent layer.
• Mentor AI and backend engineers and lead design and architecture reviews.
• Partner with US‑based AI leadership to translate strategy into well‑architected execution.

Required Skills & Qualifications

• Senior or staff‑level applied AI/ML experience shipping production systems.
• Deep fluency in modern LLM patterns: RAG, tool‑use/agents and structured output.
• Strong systems architecture skills, with the ability to design the engine, data interfaces and agent layer as a coherent whole.
• Demonstrated rigor in grounding, hallucination control and evaluation for LLM and agentic systems.
• Experience designing for reproducibility and determinism on top of probabilistic models.
• Strong software engineering in production Python, including robust, tested services.
• Understanding of why determinism and grounding matter in high‑stakes domains.
• Strong communication skills, with the ability to explain AI architecture and trade‑offs to technical and non‑technical stakeholders.

Technical Skills

• LLM & Agents: RAG, tool‑use/agents, LangGraph (or equivalent), prompt engineering, guardrails, managed LLMs (Bedrock / Vertex AI / Azure OpenAI)
• Reasoning & Data: Knowledge graph modeling, deterministic computation over structured data, vector stores (pgvector, Pinecone, Weaviate, Qdrant), SQL, Snowflake, PostgreSQL
• Platform: Python, containerized and customer‑hosted deployment, multi‑cloud model serving, observability and AI evaluation tooling

Preferred Qualifications

• Experience in finance, accounting or other high‑stakes, auditable domains.
• Knowledge graph, semantic layer or reasoning‑engine experience.
• Human‑in‑the‑loop systems and feedback loops.
• Experience with managed LLMs in‑VPC and multi‑cloud deployment.