KEY RESPONSIBILITIES
Agent design & development
Design and build agentic applications (single-agent and multi-agent) using LangChain and LangGraph with Mistral LLMs as the reasoning engine.
Model agent behaviour as LangGraph state graphs: nodes, edges, conditional routing, loops, sub-graphs, and supervisor / hierarchical multi-agent patterns.
Implement tool / function calling with Mistral models: define tool schemas, handle parallel tool calls, validate arguments and recover from tool errors.
Build Retrieval-Augmented Generation (RAG) and agentic RAG pipelines using Mistral Embed, vector stores, re-ranking and query routing.
Use structured outputs (JSON mode / schema-constrained output, Pydantic) to make agent responses reliable for downstream systems.
Add memory and persistence (LangGraph checkpointers, short- and long-term memory) and human-in-the-loop approval steps (interrupts) for high-impact actions.
Mistral platform engineering
Build agents natively on Mistral Studio where appropriate: Agents API, built-in connectors / tools, and MCP-based integrations with enterprise systems.
Orchestrate long-running, durable agentic processes with Mistral Studio Workflows (stateful execution, retries, resumption, scheduling).
Use Mistral's specialist models where they fit: Mistral OCR for document understanding, Codestral / Devstral for code tasks, Voxtral for speech, Pixtral / multimodal models for vision.
Select the right model per task (Large / Medium / Small / Ministral) balancing quality, latency, cost and data-residency requirements.
Apply fine-tuning or prompt optimisation where off-the-shelf performance is not sufficient.
Deployment & operations (LLMOps)
Package agents as APIs and services (FastAPI, LangGraph Server / Platform) and containerise them with Docker.
Deploy to cloud and on-premise targets: Mistral Studio deployments, Azure AI Foundry, AWS Bedrock, Google Vertex AI, Kubernetes, or self-hosted open-weight Mistral models served with vLLM / TGI on GPUs.
Set up CI/CD, environment and secret management, versioning of prompts, tools and graphs, and safe rollout (staging, canary, rollback).
Instrument agents with tracing and observability (LangSmith or equivalent, OpenTelemetry), and monitor latency, token cost, tool-call failures and quality drift.
Build evaluation suites: golden datasets, LLM-as-judge, task success rate, tool-use accuracy and regression tests run before every release.
Security, governance & responsible AI
Defend agents against prompt injection (direct and indirect, e.g. through retrieved documents and tool outputs), jailbreaks and data exfiltration.
Apply guardrails: Mistral moderation / guardrailing, input-output validation, least-privilege tool permissions, and PII handling.
Ensure deployments meet enterprise privacy, data-sovereignty and audit requirements; document model and data lineage.
Practice & community
Contribute reusable accelerators, reference architectures and code templates to the Mistral CoE.
Run demos and PoCs with clients and translate business problems into agent designs.
Mentor junior developers and keep current with new Mistral releases, webinars and courses on learn.mistral.ai.
CORE MISTRAL PRACTITIONER COMPETENCIES
Aligned to the technical learning tracks and levels (L100 → L400) on the Mistral learning portal.
Competency area What the practitioner must be able to do Target level
Mistral Studio – Tech - Navigate Mistral Studio; use the API and Python / TypeScript SDK; manage keys, models and usage; build and test agents with the Agents API, connectors and MCP. - L200–L300
Mistral Studio Workflows – Tech - Write workflows in code with the Mistral SDK; design durable, fault-tolerant agentic processes with retries, state and observability; run workers and deployments. -L300
Mistral Studio Deploy – Tech - Choose and execute a deployment model (managed API, cloud marketplace, VPC / on-prem, self-hosted open weights); size GPUs; secure and monitor deployments.- L300–L400
Mistral Vibe for Code – Tech -Use Mistral Vibe and Codestral / Devstral for AI-assisted development with good engineering discipline (reviews, tests, small verified changes). - L200
Mistral model family - Explain trade-offs across Mistral Large, Medium, Small, Ministral, Codestral, Mistral Embed, OCR, Voxtral and multimodal models; know open-weight vs commercial licensing.- L200
Document & multimodal AI -Build document-processing pipelines with Mistral OCR (bounding boxes, block classification, confidence scores) feeding agents and search indexes.- L200–L300
Production-ready AI systems-Move an agent from PoC to a supported live system: evaluation, guardrails, monitoring, cost control and change management.-L300–L400
TECHNICAL SKILLS
Must have
Python: strong, production-quality Python (typing, async, testing, packaging).
Mistral AI: hands-on use of the Mistral API / SDK, function calling, structured outputs, embeddings and the Agents API.
LangChain: langchain-mistralai (ChatMistralAI, MistralAIEmbeddings), LCEL, tools, retrievers, output parsers, with_structured_output.
LangGraph: StateGraph, conditional edges, checkpointers / memory, interrupts (human-in-the-loop), streaming, multi-agent supervisor patterns.
RAG: chunking strategies, vector databases (e.g. FAISS, Qdrant, pgvector, Azure AI Search, Elasticsearch), hybrid search and re-ranking.
APIs & deployment: FastAPI, REST, Docker, Git, CI/CD (GitHub Actions / Azure DevOps), at least one major cloud (Azure, AWS or GCP).
Evaluation & observability: LangSmith or equivalent tracing, building eval datasets and automated quality checks.
Good to have
Mistral Studio Workflows, MCP server / client development and enterprise connectors.
Self-hosting open-weight Mistral models with vLLM / TGI, quantisation, Kubernetes and GPU capacity planning.
Fine-tuning Mistral models (LoRA / Mistral fine-tuning API) and dataset preparation.
LangGraph Platform / Server deployment, Redis / Postgres-backed persistence.
Mistral OCR, Voxtral (speech) or multimodal use cases.
Security background in LLM threats (OWASP Top 10 for LLM Applications, prompt-injection testing).
TypeScript / LangChain.js for front-end or Node-based agent services.
QUALIFICATIONS
Bachelor's or Master's degree in Computer Science, Engineering, Data Science or related field (or equivalent experience).
Demonstrable portfolio: at least one agentic application built with LangChain / LangGraph and deployed beyond a notebook (GitHub, demo or case study).
Completion of the Mistral learning-portal technical tracks (Studio, Workflows, Deploy), or willingness to complete them within the first 60 days.
Clear communication with technical and business stakeholders; comfortable presenting demos to clients.
WHAT SUCCESS LOOKS LIKE (FIRST 90 DAYS)
Timeline Outcomes
30 days - Completes Mistral Studio and Workflows technical courses (L100–L200). Builds a LangGraph + Mistral reference agent with tool calling, RAG and tracing.
60 days - Completes the Studio Deploy track. Deploys the reference agent to a staging environment with CI/CD, evaluation suite and guardrails.
90 days - Delivers or co-delivers a client PoC / internal use case to production-ready standard and contributes a reusable accelerator to the CoE.
SUGGESTED ASSESSMENT
Technical screen: Mistral model selection, function calling, RAG design, LangGraph state and memory concepts.
Hands-on exercise (2–3 hrs): build a LangGraph agent on a Mistral model that uses two tools, returns structured output and includes a human-approval step; expose it via FastAPI in Docker.
System design: architect a production deployment covering hosting choice, scaling, observability, evaluation, security and cost.
Practitioner check: review of Mistral learning-portal course completions and badges.
Reference: Mistral AI learning portal – https://learn.mistral.ai/home/ (technical tracks: Mistral Studio, Studio Workflows, Studio Deploy, Mistral Vibe for Code; levels L100–L400). Bracketed fields are placeholders to fill in.
EY | Building a better working world
EY exists to build a better working world, helping to create long-term value for clients, people and society and build trust in the capital markets.
Enabled by data and technology, diverse EY teams in over 150 countries provide trust through assurance and help clients grow, transform and operate.
Working across assurance, consulting, law, strategy, tax and transactions, EY teams ask better questions to find new answers for the complex issues facing our world today.
Apply now »
Apply now
Start applying with LinkedIn
Apply Now
Start
Please wait...
Find similar jobs
- Consulting DE,
- Consulting CN,
- Consulting CA,
- Consulting BR,
- Consulting
Connect with us
Our locations
My EY
Site map
Legal and privacy
Labor condition applications
Cookie Consent Manager
Opens in a new tab.
Opens in a new tab.
Opens in a new tab.
Opens in a new tab.
Opens in a new tab.
EY refers to the global organization, and may refer to one or more, of the member firms of Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global Limited, a UK company limited by guarantee, does not provide services to clients.
×
Cookie Consent Manager
When you visit any website, it may store or retrieve information on your browser, mostly in the form of cookies. Because we respect your right to privacy, you can choose not to allow some types of cookies. However, blocking some types of cookies may impact your experience of the site and the services we are able to offer.
Required Cookies
These cookies are required to use this website and can't be turned off.
Required Cookies
Show More Details
Required Cookies
- Provider
- Description
- Enabled
SAP as service provider
We use the following session cookies, which are all required to enable the website to function:
"route" is used for session stickiness
"careerSiteCompanyId" is used to send the request to the correct data center
"JSESSIONID" is placed on the visitor's device during the session so the server can identify the visitor
"Load balancer cookie" (actual cookie name may vary) prevents a visitor from bouncing from one instance to another
Cookies from provider SAPasserviceprovider are required and cannot be turned off
Advertising Cookies
These cookies serve ads that are relevant to your interests. You may freely choose to accept or decline these cookies at any time. Note that certain functionality that these third parties make available may be impacted if you do not accept these cookies.
Consent to all Advertising Cookies
Show More Details
Advertising Cookies
- Provider
- Description
- Enabled
AddThis
Google Analytics is a web analytics service offered by Google that tracks and reports website traffic.
Cookie Information
Privacy Policy
Terms and Conditions
Consent to cookies from provider AddThis
LinkedIn
LinkedIn is an employment-oriented social networking service. We use the Apply with LinkedIn feature to allow you to apply for jobs using your LinkedIn profile. Opting out of LinkedIn cookies will disable your ability to use Apply with LinkedIn.
Cookie Policy
Cookie Table
Privacy Policy
Terms and Conditions
Consent to cookies from provider LinkedIn
Google Analytics
Google Analytics is a web analytics service offered by Google that tracks and reports website traffic.
Cookie Information
Privacy Policy
Terms and Conditions
Consent to cookies from provider GoogleAnalytics
Google Tag Manager
Google Tag Manager is a tag management system for conversion tracking, site analytics, remarketing, and more.
Privacy Policy
Terms and Conditions
Consent to cookies from provider GoogleTagManager
Confirm My Choices
- Reject All Cookies
- Accept All Cookies