Staff Data Scientist - Agentic AI

A Staff Data Scientist will be a hands-on builder of production AI agents - not a researcher who experiments with agent frameworks on the side. You will be a key member of our fast-growing, high-performing team in India, owning the components that turn an LLM call into a reliable, autonomous system that can reason, act, and recover in production.

Responsibilities:

• Agent Architecture and Technical Strategy: Define the roadmap for our agentic AI stack => the execution loop, harness, memory, and tool/skill layers. And decide when an agent, a classical model, or a hybrid is the right tool for a given revenue problem.
• Production Agent Engineering: Build and operate the core components of our agents end-to-end: the reasoning/execution loop, the harness that manages tool calls, retries, timeouts and session state, short- and long-term memory, and the skill/tool registry agents draw on (via MCP-style tool calling).
• Planning and Multi-Step Reasoning: Design task-decomposition and planning strategies (evidence-based planning, plan-execute, multi-hop reasoning, research and many more) so agents can coach sellers, inspect deals, raise forecast risks and update CRMs autonomously and correctly.
• Guardrails, Trust and Evaluation: Own the evaluation framework for agentic behaviour - offline eval, LLM-as-judge, and online A/B testing plus the guardrails (input/output validation, policy and safety checks) that keep agents reliable at enterprise scale.
• GenAI and Revenue Modelling: Apply rigorous statistical and time-series methods to our core revenue models (Forecasting, Deal Health, and Risk Prediction) and connect them into agentic workflows where appropriate.
• Multi-Agent and Cross-System Coordination: Design how agents talk to sub-agents and other systems (agent-to-agent-style communication and tool/skill registries shared across agents) so capability is composed rather than rebuilt per use case.
• Cross-Functional Technical Leadership: Partner with product and engineering leadership to translate business objectives into concrete agent designs and serve as the technical anchor who can explain agent behaviour and failure modes to non-technical stakeholders.
• Mentorship and Culture: Mentor senior data scientists on agent-building discipline, not just prompting, and foster a culture of rapid experimentation paired with production rigour.
• Enablement: Contribute to internal documentation, onboarding, and training on our agent components and patterns, promoting platform adoption across teams.

Requirements:

• Experience: 8+ years in data science or machine learning, of which at least 2-3 years must be hands-on building and operating production AI agents (staff/lead experience preferred).
• Core ML and Stats: Deep expertise in Classical ML (XGBoost, Causal Inference), Time-Series Forecasting, and Deep Learning fundamentals - you understand the maths behind the models, not just how to import libraries.
• Generative AI Foundations: Proficiency with modern agent/LLM frameworks (LangChain, LlamaIndex, DSPy, or equivalent in-house harnesses), vector databases (Pinecone, Elasticsearch), and techniques like fine-tuning (PEFT/LoRA) and RAG optimisation.
• Engineering First: Expert-level Python and SQL. You write production-ready code (modular, tested, typed) and work comfortably within a larger software engineering ecosystem (Git, Docker, CI/CD).
• System Design for Agents: Ability to design end-to-end agentic systems and articulate trade-offs between reasoning depth, latency, cost per agent turn, and reliability - and make architectural decisions that hold up at enterprise scale.
• Education: An MS or PhD in computer science, statistics, physics, maths, or an equivalent quantitative field is preferred.

Agentic AI - Component Depth: Practical production experience across most of the following

• Agent loop / orchestration - ReAct-style or plan-execute reasoning loops.
• Agent harness: tool-calling interfaces, retry and timeout handling, sandboxing, session/state management.
• Memory: short-term context management and long-term memory (vector- or graph-backed, episodic).
• Tools and skills: function calling, MCP-style tool registries, and skill libraries agents can discover and invoke.
• Planning: task decomposition, multi-step and evidence-based planning.
• Retrieval and context engineering: RAG, Graph RAG, and context assembly for grounded reasoning.
• Guardrails and safety: input/output validation, policy enforcement.
• Evaluation and observability: LLM-as-judge, online A/B testing, tracing.
• Multi-agent coordination: Agent-to-Agent (A2A)-style communication and sub-agent delegation.