Description

1. Develop comprehensive red-team strategies for any agentic Al application, covering adversarial, boundary, misuse, ambiguity, robustness, privacy, compliance, and business-logic testing across customer search, lead generation, coverage-gap analysis, policy servicing, product information, incentive campaigns, and sales-support use cases.
2. Design high-quality single-turn and multi-turn adversarial prompts that test direct requests, indirect requests, role-playing, context manipulation, prompt injection, instruction conflicts, progressive disclosure, cross-turn dependency, memory contamination, follow-up ambiguity, and attempts to override application or policy guardrails.
3. Possess solid insurance-domain knowledge, preferably within life insurance, health and protection products, retirement planning, wealth accumulation, critical illness, policy maturity, premium cessation, lapsed policies, customer segmentation, coverage-gap assessment, financial-needs analysis, and agent incentive campaigns. The incumbent should understand how apparently valid insurance queries can create suitability, mis selling, privacy, discrimination, or regulatory risks.
4. Create realistic and risk-focused test scenarios for insurance advisers and distribution users, including customer prioritization, lead identification, demographic filtering, high-net-worth segmentation, policy recommendations, cross-selling opportunities, policy events, incentive qualification, and customer outreach. Test cases must distinguish between legitimate business assistance and unsupported or potentially non-compliant recommendations.
5. Evaluate the complete agentic workflow rather than only the final natural-language response, including intent classification, query sanitization, routing, tool selection, function parameters, data retrieval, ranking logic, structured-data generation, follow-up suggestions, refusal behavior, fallback handling, and consistency between textual and UI outputs.
6. Define precise expected behavior, expected safe behavior, expected intent, assertion rules,



and scoring rubrics for each test case. The incumbent must convert insurance requirements and control clauses into measurable acceptance criteria that can determine whether a response is correct, grounded, complete, safe, compliant, relevant, and appropriately qualified.
7. Design, configure, and calibrate Judge LLM evaluation frameworks for automated assessment of Agentic AI application responses. This includes developing judge prompts and rubrics, defining pass/fail and severity thresholds, measuring factuality and groundedness, testing judge consistency, identifying false positives and false negatives, managing positional or verbosity bias, and validating the Judge LLM against expert human annotations.
8. Test hallucination resistance and grounding quality, particularly where insurance product details, campaign rules, customer records, policy attributes, or eligibility criteria are missing, incomplete, contradictory, or outside the application's supported data sources. The incumbent should verify that the agent does not fabricate customer facts, campaign thresholds, policy benefits, product suitability, or qualification outcomes, and that it provides an appropriate limitation statement or safe alternative.
9. Assess privacy, confidentiality, fairness, and data-minimization controls when working with sensitive customer information such as age, income, occupation, marital status, policy holdings, coverage gaps, residency indicators, and financial circumstances. Testing should identify over-disclosure, unauthorized profiling, discriminatory filtering, improper customer ranking, exposure of personal data, and unsafe use of protected or sensitive attributes.
10.



Validate insurance compliance and conduct-risk guardrails, including product-suitability boundaries, financial-needs-analysis requirements, non-solicitation and cross-border considerations, unsupported financial recommendations, unfair customer targeting, misleading comparisons, and the use of appropriate disclosures.
11. Apply strong technical skills in LLM and Agentic Al testing, including prompt engineering, retrieval-augmented generation, tool/function calling, agent orchestration, context-window behavior, conversation-state management, JSON and JSONL schemas, API and streaming-response analysis, structured-output validation, automated assertions, Python-based testing, and interpretation of latency, status codes, execution results, conversation IDs, and raw model responses.

Analyze red-team execution results and drive remediation with engineering, product, compliance, data, and business teams. The incumbent should classify failures by severity and business risk, distinguish model failures from data, orchestration, retrieval, UI, or Judge LLM failures, identify recurring patterns, propose guardrail or prompt improvements, maintain regression suites, produce clear release-readiness reports with evidence, reproducible test cases, residual risks, and recommended corrective actions.

Enable Skills-Based Hiring No ECMS REQ IDAdditional Details

- ECMS REQ ID : 580978
- Project Location 1 : KRNK | BANGALORE
- Project Location 2 : (No Value)
- Project Location 3 : (No Value)
- Project Location 4 : (No Value)
- Delivery SPOC : NA
- Relevant Experience : 6+
- Mandatory skills : Strong hands on with Red Teaming, Agentic AI and LLM Applications experience.
- Desired skills : Security Testing certifications
- Domain (Industry) : secuirty
- BGC (Before onboarding / After onboarding) : PreOB
- Total Experience (Ex. 5-7 Years) : 8+
- BGC Details : pre bgv
- BGC Vendor : nascom
- Mode of Interview : Face to Face
- WFO / WFH / Hybrid : Hybrid
- Please enter shift timings : 9-6
- Shift Timings : General

📌 580978-Red Teaming, Agentic AI and LLM Applications experience (India)
🏢 Trigent Software
📍 India