Tecafe - AI Consultant - LLM/RAG

Description :

- Seniority Senior Consultant (8 to 12 years total; 2 to 3+ years in GenAI /agentic systems)

- Shift- 6 : 00 PM to 3 : 00 AM

- Role Type- Contract (1 years-Extendable)

- Location -Remote

- Start Date -Immediate

1. Key Deliverables :

The consultant will be accountable for the following four deliverables. Each deliverable must be

evidenced by working artifacts (designs, code, evaluation results) and measurable outcomes.

AI and Agentic Capability Enhancement :

Design, develop, and enhance AI-driven capabilities within the existing platform leveraging LLM-

based and agentic system components while preserving the current multi-agent architecture.

- Extend existing agents (and, where justified, introduce new specialist agents) within the current

orchestration pattern; no rearchitecture of the multi-agent foundation unless explicitly

approved.

- Improve prompting strategies, tool-calling, retrieval (RAG), memory, and reflection/critique loops

to raise reasoning quality and consistency.

- Harden agent behavior through evaluation harnesses, regression tests, guardrails, and

observability (traceability of reasoning, safety checks, hallucination detection).

- Document all enhancements as production-ready design specs, prompt/behavior blueprints, and

runbooks.

RCA and Recommendation Simplification :

- Improve the output of RCA findings and Next Best Action recommendations to reduce excessive

technical verbosity while maintaining analytical accuracy and decision confidence.

- Redesign the RCA narrative structure so outputs lead with the verdict, confidence, and

recommended action, followed by progressively disclosed technical evidence.

- Introduce length, tone, and density controls at generation time (not just post-hoc summarization)

so simplification does not degrade fidelity.

- Apply RCA methodologies (e.g., 5 Whys, Ishikawa, fault tree) inside the agentic flow so that

simplified outputs remain auditable and traceable back to underlying evidence.

- Define and track quality metrics : diagnostic accuracy, time-to-root-cause, analyst acceptance

rate, verbosity/signal ratio, and recommendation confidence calibration.

Persona-Aware Output Adaptation :

- Apply persona-aware output adaptation so that the same underlying RCA can be presented differently

based on the user's role, required level of detail, and presentation format. In-scope personas include :

- Frontline agents concise, action-first, customer-safe language; minimal jargon; clear next

step.

- Supervisors case-level summary with confidence, exception flags, team/queue implications,

and escalation triggers.

- Advanced technical support full technical depth, evidence chain, telemetry references,

hypothesis trail, and diagnostic artifacts.

Specific expectations :

- Build a persona catalog covering tone, authority, decision boundaries, terminology, and

escalation thresholds for each in-scope persona.

- Operationalize persona selection and adaptation within the agent flow (system prompts, role-

conditioned generation, formatting templates) not as a separate downstream rewrite step.

- Establish governance for persona changes : versioning, approval workflow, A/B testing, and

rollback.

- Validate persona adherence through human-in-the-loop review and automated persona-conformance checks.

Workflow and Interface Integration :

- Integrate all enhancements into existing workflows and user interfaces, supporting both live

troubleshooting scenarios and post-ticket analysis use cases.

- Integrate with existing UI surfaces and workflow tools used by frontline, supervisor, and

technical support personas without requiring users to change tools.

- Support real-time/low-latency delivery for live troubleshooting and batch/analytical delivery for

post-ticket analysis.

- Integrate with enterprise data sources and tooling (telemetry, logs, knowledge bases, ticketing)

through approved interfaces.

- Provide hypercare support post-launch : monitoring, defect triage, prompt/persona tuning, and

stakeholder enablement.

2. Key Responsibilities :

Delivery & Stakeholder Management :

- Lead the enhancement workstream end-to-end : discovery, design, build, pilot, hardening, and hypercare.

- Partner with product, data, operations, risk, and compliance teams to align agent behavior and personas to business KPIs, SLAs, and risk tolerances.

- Define, baseline, and track success metrics (e.g., MTTR, time-to-root-cause, diagnostic accuracy, analyst acceptance, verbosity reduction, persona conformance).

- Present design decisions, pilot outcomes, and roadmaps to senior stakeholders in clear, business-friendly language.

Personality & Persona Engineering :

- Translate business intents, user personas, and RCA objectives into concrete agent architectures and personality designs.

- Produce personality and behavior blueprints communication style, decision boundaries, escalation thresholds, and safety constraints for each in-scope persona.

- Operationalize prompting strategies (system prompts, tool instructions, reflection loops) so agents behave consistently and safely across channels.

Multi-Agent GenAI & RCA Engineering :

- Enhance the existing multi-agent system signal triage, hypothesis generation, evidence gathering, recommendation synthesis, verification preserving the established architecture.

- Implement evaluation and observability : hallucination detection, safety checks, reasoning traceability, and persona adherence.

- Ensure generated root-cause narratives and NBAs are auditable, explainable, and actionable for the target persona.

3. Required Experience & Proof Points :

Candidates must demonstrate hands-on experience supported by concrete proof points case

studies, references, portfolio artifacts, or sanitized code/design samples.

GenAI & Multi-Agent Systems :

- 5 to 8+ years in AI/ML; 23+ years focused on GenAI and agentic systems.

- Shipped at least one production or late-stage pilot of a multi-agent GenAI system (e.g., orchestrated LLM agents for support, operations, or engineering workflows).

- Demonstrable experience with at least one agentic framework (LangChain / LangGraph, CrewAI, AutoGen, or equivalent) and with tool-calling / RAG patterns.

- Strong applied knowledge of LLM evaluation curated test sets, human-in-the-loop review, regression testing across prompt/persona/tool changes.

- Evaluation & Observability for LLM Agents.

Personality & Persona Engineering :

- Built persona-based assistants or role-specialized agents in an enterprise setting (e.g., ops copilot, contact-center coach, network/RCA assistant).

- Tangible persona artifacts : system prompt specifications, persona catalogs, behavior taxonomies, agent charters, or style guides.

- Experience designing tone, authority, and decision boundaries for agents serving multiple stakeholder personas (frontline, supervisors, engineers, execs).

- AI Safety & Policy-Bound Behavior.

Root Cause Analysis & Observability :

- Hands-on RCA experience in complex systems (network, IT ops, customer journeys, process failures) ideally with automation/AI in the loop.

- Experience integrating AI agents with monitoring/observability (logs, metrics, traces, events) and ticketing/workflow tools.

- Documented improvements in MTTR, defect recurrence, or diagnostic throughput via AI/automation (with baselines and outcomes, even if anonymized).

Enterprise Delivery & Self-Management :

- Track record of owning a workstream inside a larger program planning, stakeholder management, risk management, status reporting.

- Experience in regulated or high-oversight environments (telecom, financial services, healthcare,

or large-scale consumer platforms).

- Strong communication skills : running design sessions, show-and-tells, and enablement

workshops with mixed technical and non-technical audiences.

4. Core Technical Skills & Competencies :

LLMs & GenAI :

- Prompt and system-instruction design; role-conditioned generation.

- RAG architectures and retrieval design.

- Tool calling, workflow orchestration, and structured outputs.

- Multi-agent patterns : plannerworker, supervisorspecialist, debate/critique loops.

Platforms & Languages :

- Experience with one or more cloud AI platforms (Azure OpenAI, AWS Bedrock, Google Vertex, Anthropic APIs).

- Proficient in Python (preferred) and relevant libraries/frameworks for agentic systems, evaluation, and integration.

AI Safety, Security & Governance :

- Guardrails, PII handling, access control, auditability, and responsible AI practices suitable for an enterprise program.

- AI Safety & Policy-Bound Behavior

RCA & Systems Thinking :

- Solid grounding in RCA methodologies, systems thinking, and complex incident analysis.

- Ability to translate RCA patterns into agent behaviors and workflows triage, hypothesis generation, evidence gathering, recommendation, verification.

Consulting & Collaboration :

- Problem framing, hypothesis-driven analysis, stakeholder alignment, and storytelling with data.

- Self-managing delivery : sprint planning, roadmap ownership, high-quality documentation.

5. Preferred Profile :

- 8 to 12 years total experience in AI/ML, software engineering, or data science, with 2 to 3+ years on GenAI and agentic systems.

- Prior experience in telecom, large-scale consumer tech, or similarly complex operational environments.

- Experience working inside or alongside enterprise AI/ML platforms, AI governance boards, or innovation labs.

- Prior exposure to agentic RAG RCA solutions and multi-agent orchestration for complex operational workflows.

7. Supplier Submission Requirements :

For each candidate proposed, the supplier should provide :

- Updated CV with relevant GenAI / agentic / RCA engagements highlighted.

- Short summary (1-21 page) mapping the candidate's experience to the four Key Deliverables in Section 2.

- At least two references or case studies covering multi-agent GenAI and persona-based solutioning.

- Daily/weekly rate, earliest available start date, notice period, and location/working-mode constraints.

- Confirmation of the candidate's ability to comply with the client's security, data handling, and background-check requirements.