We are building an agentic CRM migration platform that helps consultants analyze, compare, and plan migrations between enterprise CRM systems (Salesforce, Microsoft Dynamics). The platform combines LLM-powered code analysis, knowledge graphs, RAG-based conversational AI, and structured data pipelines to automate what was previously weeks of manual discovery work.

The stack is production-grade:
• Python/FastAPI backend
• LangGraph multi-agent orchestration
• Neo4j dependency graphs
• PostgreSQL + pgvector for hybrid retrieval
• Azure OpenAI for LLM inference
• React frontend
• Deployed on Azure (AKS)

We already have a working platform with codebase parsing, graph visualization, process map generation, and a migration planner agent. You would join a small, high-velocity team to scale the AI layer across multiple new features shipping in parallel.

What You'll Own

Multi-agent systems:
Design, build, and maintain LangGraph ReAct agents that orchestrate tools across knowledge graphs, vector search, and relational data.
Current agents include:
• Graph QA agent
• Process map HITL agent
• Migration planner
You’ll build the platform’s global AI assistant and extend existing agents.

RAG pipeline architecture:
Own the end-to-end retrieval pipeline:
• Embedding strategy (OpenAI text-embedding-3-small)
• Hybrid search (pgvector dense + PostgreSQL full-text)
• Reciprocal Rank Fusion
• Optional reranking (Cohere/Bedrock)
Tune retrieval quality across code analysis, process maps, and aggregate analysis content.

Prompt engineering & evaluation:
• Author and maintain system prompts for code analysis, entity/field comparison, capability classification, R/R/R triage integration, and aggregate cross-file synthesis
• Build evaluation harnesses to measure prompt quality against ground truth

LLM-powered data pipelines:
• Design batch LLM workflows for data model comparison (static YAML + LLM hybrid matching)
• Cross-file aggregate analysis (graph-guided context assembly with tiered summarization)
• Capability classification
• Enforce concurrency controls, token budgets, and graceful degradation

Graph + AI integration:
Work closely with Neo4j graph data (Cypher queries, property enrichment, subgraph extraction) to ground AI outputs in structured CRM knowledge. Ensure efficient graph querying and proper data flow into downstream AI features.

Architecture & technical leadership:
• Contribute to work breakdowns, PR strategies, and cross-feature dependency planning
• Review AI-related PRs from backend developers
• Mentor the team on LLM best practices, agent patterns, and retrieval tuning

Required Skills & Experience:

Core (must-have)
• 5+ years in applied ML/AI engineering, with 2+ years building LLM-powered production systems
• Multi-agent orchestration (LangChain, LangGraph, or similar)
• RAG systems (embeddings, vector DBs, hybrid retrieval, reranking, chunking)
• Python (async, FastAPI, production services)
• Prompt engineering (classification, extraction, synthesis)
• LLM evaluation (LLM-as-judge, scoring, eval pipelines)
• OpenAI API (chat, embeddings, rate limits, batch processing)

Strongly preferred:
• Graph databases (Neo4j, Cypher)
• Azure OpenAI (deployments, Managed Identity, AKS)
• CRM domain knowledge (Salesforce / Microsoft Dynamics)
• PostgreSQL + pgvector
• Evaluation frameworks (precision/recall, HITL loops)
• Knowledge graph enrichment with LLMs
• Production AI observability (cost, latency, quality monitoring)

Nice to have:
• AI safety & compliance (PII, prompt injection, audit trails)
• PhD or advanced degree in a quantitative field
• Cohere or cross-encoder reranking experience
• Background in code analysis or developer tooling

Team Context:

You’d be the second AI Architect on the team, working alongside an existing AI Architect with deep experience in multi-agent platforms, structured-data RAG, and evaluation pipelines.

The broader team includes:
• Backend developers
• Frontend developers
• CRM domain experts

You’ll collaborate closely with backend engineers on integration and with the CRM expert on domain-specific prompts.

Key Deliverables (First 3 Months):

• Global AI Assistant (LangGraph ReAct agent with hybrid RAG + graph tools)
• Aggregate code analysis prompts (cross-file synthesis)
• Data model comparison prompts (entity/field matching)
• R/R/R triage integration into migration planner
• Retrieval quality tuning with measurable improvements

Additional Info:

• Start date: ASAP
• HackerRank Challenge: Yes
• Fully remote, with occasional in-person sessions (≈1x per quarter, Prague)
• US hours overlap required: 2–6 PM CET

‍