Senior Solution Architect

AI agents are only as good as the context they can reach. Most CX questions are simple lookups; some need policy and knowledge content, and a few need to know who the customer really is, how their accounts connect, and which scenario they're in. The MCP Gateway gives agents one interface for all of it. You'll own what sits behind that interface.

You'll design and build Synqed's context graph and hybrid retrieval platform: the system that powers entity resolution, scenario routing, multi-hop reasoning, and permission-aware retrieval for enterprise CX agents. This is not a slide-ware architecture role. You'll write the graph queries, build the pipelines, and ship the code that agents depend on in real time.

The core responsibilities for the job include the following:

Build the context graph:

• Design the CX-specific canonical schema: canonical customer entities, SAME_AS edges to source-system IDs, household and account hierarchies, scenario nodes, and bitemporal edges.
• Build and run a production graph database (Neo4j, TigerGraph, Neptune, or similar) with B2B2C multi-tenant isolation.
• Integrate an entity resolution engine (Senzing or Tilores) and own the CX-aware layer on top of it.
• Model and query the relationships that matter: households, account hierarchies, scenario-to-tool mappings, invocation logs, and permission graphs.

Unify retrieval behind one interface:

• Architect the retrieval layer that routes each query to structured APIs, vector search, or graph traversal based on its shape.
• Build and tune hybrid search pipelines on pgvector, Qdrant, or similar, combining vectors, BM25, and reranking (Cohere, BGE).
• Build the graph-backed skill router: traverse the scenario subgraph from a conversation's context, pick the right CX skill, and load its compiled IR.
• Hit the latency targets in every mode, and keep hitting them as tenants and data grow.

Get enterprise data in, and keep it fresh:

• Build ingestion pipelines from structured, semi-structured, and unstructured sources: Salesforce, HubSpot, Zendesk, SAP, and custom APIs.
• Implement change data capture with Estuary Flow and native SaaS APIs (Salesforce Pub/Sub, HubSpot webhooks, Zendesk audit logs), using a uniform ChangeEvent envelope mapped to the canonical schema.
• Design tenant onboarding: seed systems of record into the graph, the vector index, and live-query caches.

Put it to work for AI agents:

• Build Graph RAG that combines graph traversal, vector search, and LLM reasoning for multi-hop CX queries.
• Build agent workflows with LangChain, LangGraph, Anthropic tool use, CrewAI, or a similar tool, integrated with the MCP Gateway.
• Design the eval suite that runs alongside every scenario and checks agent behavior against customer-owned policy.

Run it in production, and in front of customers:

• Own the system-level calls: query optimization, indexing, traversal depth limits, caching, and performance tuning.
• Ship with proper observability, versioning, and rollback.
• Work forward-deployed: embed with customers to map their data, co-build flows and ship the runtime, then present the design to their architects with clarity and confidence.

Requirements:

• You've built it, not just drawn it. 4-8 years building production systems, including graph-based or retrieval systems you designed and shipped yourself. You write graph queries, debug traversal performance, review pipelines, and ship code.
• Graph depth. Hands-on with Neo4j, TigerGraph, Amazon Neptune, or similar. You know graph modelling inside out (nodes, relationships, properties, labels, directionality, traversal patterns, bitemporal edges), and you're fluent in Cypher, Gremlin, or equivalent, including query, index, and traversal optimization.
• LLM and agentic systems. You've built with function calling, tool use, and retrieval-augmented generation, using LangChain, LangGraph, Anthropic tool use, CrewAI, or similar. You understand embeddings, similarity search, and hybrid retrieval with reranking.
• Data engineering. You've built ETL/ELT pipelines, CDC streams, and cross-system integrations at scale, and you've worked with entity extraction, entity resolution, or record linkage.
• Systems fundamentals. Strong system design and distributed systems thinking: scalability, fault tolerance, consistency models, and caching.
• Polyglot and production-ready. Comfortable across Python, TypeScript/Node.js, and SQL, with containerized deployments on Docker and Kubernetes.
• A builder who gets things done. You do your best work from a blank page: you figure it out, try things fast, and get them done without waiting to be told. No task is above or below you.
• Location. Based in, or willing to relocate to, Bangalore. Remote is possible for exceptional candidates.

Even Better:

• Graph RAG in production: Hybrid retrieval architectures, or retrieval systems built for LLM-powered products.
• Entity resolution platforms: Senzing, Tilores, or equivalent.
• Neo4j beyond the basics: APOC, full-text search, vector indexes, or graph analytics libraries.
• Streaming and CDC: Debezium or Estuary Flow, and event streaming with Kafka, SQS, or Pub/Sub.
• Enterprise SaaS data models: You know the APIs and data models of Salesforce, HubSpot, SAP, or Zendesk.
• CX and multi-tenancy: CX workflows, contact center operations, or B2B2C multi-tenant architectures.
• MCP: Model Context Protocol or similar agent-to-tool protocols.