AI Model Governance Specialist (Risk)

ICICI Securities; Navi Mumbai; On-site; INR 3,000,000 - 6,000,000; Full time

Job summary

ICICI Securities seeks a senior specialist to independently evaluate AI/LLM systems, design robust evaluation frameworks, and establish controls for secure enterprise AI deployment.

The role covers end-to-end RAG pipelines, risk vectors like data leakage and prompt injection, and governance with pre/post deployment validation, observability, and continuous drift monitoring. 9–12 years of relevant experience is preferred.

Qualifications

• Strong quantitative/CS foundation.
• Deep understanding of AI/GenAI workflows.
• Familiarity with RAG architecture and risks.
• Experience with reference-based testing and LLM evaluation.

Responsibilities

• Independently evaluate AI/LLM systems and establish controls for secure enterprise deployment.
• Conduct RAG & GenAI risk assessments across end-to-end pipelines.
• Design evaluation frameworks to measure hallucination, bias, robustness, and calibration.
• Curate datasets and test cases; validate LLM outputs and governance controls.
• Support enterprise AI control frameworks including data provenance and drift monitoring.

Skills

GenAI

LLMs

RAG

AI governance

risk assessment

NLP evaluation

Python

Education

B.Tech / M.Tech / MCA

Job description

The role will be responsible for independently evaluating AI/LLM systems, identifying potential risk vectors, designing robust evaluation frameworks, and establishing controls to support responsible and secure enterprise AI deployment.

• Conduct RAG & GenAI risk assessments across end-to-end RAG pipelines, including knowledge-base ingestion, chunking, embeddings, vector retrieval, context grounding, and generation.
• Identify AI/LLM risks such as data leakage, prompt injection, context contamination, retrieval failures, and hallucinations.
• Design and execute LLM evaluation and benchmarking frameworks to assess hallucination rates, toxicity, robustness, model calibration, fairness, and bias.
• Curate golden/reference datasets and design evaluation test cases with reference answers.
• Implement and validate LLM-as-a-Judge evaluation pipelines.
• Establish AI governance and risk mitigation controls covering data preprocessing, pre-deployment validation gates, and post-deployment observability.
• Independently inspect, evaluate, and validate AI/LLM model outputs.
• Conduct technical audits of AI/LLM systems and assess model performance against defined evaluation criteria.
• Support the development of enterprise AI control frameworks covering data provenance, model sign-off, rate-limiting, and continuous drift monitoring.
• Evaluate automated AI assessment and hallucination detection tooling, including Ragas, TruLens, and DeepEval.

Qualifications

• B.Tech / M.Tech in Computer Science, Quantitative disciplines, or MCA.
• Strong quantitative and/or Computer Science foundation.
• Thorough understanding of AI/ML taxonomy and modern Generative AI workflows.
• Deep understanding of RAG architecture and associated risks, including vector search, embeddings, relevance scoring, and context grounding/grounding boundaries.
• Hands-on familiarity with reference-based testing, LLM-as-a-Judge frameworks, red-teaming fundamentals, and semantic similarity evaluation.

Experience & Skills

9 to 12 years of relevant experience

Required Technical Skills:

• Strong expertise in Generative AI / GenAI and Large Language Models (LLMs).
• Strong understanding of Retrieval-Augmented Generation (RAG) pipelines.
• Experience in LLM evaluation, model validation, AI risk assessment, and AI governance.
• Understanding of AI risk vectors including prompt injection, hallucination, data leakage, and retrieval failures.
• Familiarity with enterprise LLM orchestration and deployment platforms such as Amazon Bedrock, Azure OpenAI Service, and Vertex AI.

Preferred / Good-to-Have Skills:

• Working knowledge of Python for querying APIs, parsing JSON outputs, and independently auditing model evaluation scripts.
• Practical experience with NLP evaluation metrics such as ROUGE, BLEU, BERTScore, and semantic distance metrics.
• Ability to construct enterprise AI control frameworks covering data provenance, model sign-off, rate-limiting, and continuous drift monitoring.
• Familiarity with automated hallucination detection and AI evaluation tools such as Ragas, TruLens, and DeepEval.