Principal Engineer — Aiva Platform

You will own the end-to-end architecture of Aiva — Aivar's memory and context graph platform — from data model to APIs to production SLOs. This is a hands-on, code-first principal role: you will set the technical direction, make the hard trade-offs (graph vs. vector vs. hybrid, consolidation vs. recall latency, tenancy isolation vs. cross-agent sharing), and build the first versions of the hardest components yourself while mentoring a small, high-leverage team. Aiva must serve three very different consumers — voice agents that need sub-100ms context recall, long-running process agents that need transactional state, and fine-tuned small models that need curated training context — behind one coherent platform, governed by the Aivar AI Gateway. Key Responsibilities - Own the Aiva platform architecture: context graph data model (entities, relationships, episodic and semantic memory layers), storage topology, and the memory lifecycle — capture, extraction, consolidation, retrieval, decay, and deletion. - Design and build the hybrid retrieval layer combining graph traversal, vector similarity, and structured filters, with hard latency budgets for real-time voice(Convogent) and throughput targets for batch agentic workloads (Velogent). - Define Aiva's API and SDK surface — including MCP-compatible interfaces — so accelerator teams and enterprise customers can read/write memory with clear contracts, versioning, and backward compatibility. - Architect multi-tenancy, data isolation, encryption, and residency controls suitable for regulated enterprise customers (BFSI, healthcare, telecom), including deployment into customer VPCs and sovereign/air-gapped environments. - Integrate memory governance with the Aivar’s Convogent, Velogent and Aivar AI Gateway and ReVAct: memory access policies, audit trails, PII redaction and consent-aware retention, and per-tenant observability and cost attribution. - Set and own production SLOs (latency, recall quality, availability) and the operational model — capacity planning, cost-per-tenant economics, and incident response. - Make build/buy/adopt decisions across the stack (e.g., Neptune vs. Neo4j vs. custom, OpenSearch vs. dedicated vector stores, event streaming choices) with clear written rationale. - Hire, mentor, and technically lead the Aiva team; establish engineering standards, design review cadence, and documentation culture. - Partner with the co-founders on Aiva's roadmap and represent Aiva's architecture in customer and partner conversations (CXO and enterprise-architect audiences). Must-Have Qualifications - 12+ years building and shipping backend/data platforms, with demonstrated principal-level ownership of at least one platform used by multiple product teams or external customers. - Deep, hands-on experience with at least two of: graph databases (Neptune, Neo4j), vector search (OpenSearch/Elasticsearch k-NN, pgvector, Pinecone, Milvus), large-scale key-value/document stores (DynamoDB, Cassandra), and event streaming (Kafka/Kinesis). - Strong distributed-systems fundamentals: consistency models, partitioning, caching strategies, backpressure, and designing for p99 latency targets. - Working knowledge of LLM application architecture: RAG, embeddings, context-window management, agent frameworks, and evaluation of retrieval quality. - Production AWS depth — the Aivar platform is AWS-native (Bedrock, EKS/Lambda, Neptune/OpenSearch, DynamoDB, S3, KMS, PrivateLink). - Proven security and compliance literacy for enterprise data platforms: encryption at rest/in transit, tenancy isolation, audit logging, and data-retention policy enforcement. - Excellent written communication — design docs, ADRs, and customer-facing architecture narratives. Nice-to-Have - Prior work on memory systems for AI agents, personalization platforms, customer data platforms (CDPs), identity/entity resolution, or feature stores. - Experience with knowledge-graph construction from unstructured data (NER, relation extraction, entity linking). - Exposure to Anthropic/Claude, Bedrock agent tooling, or the Model Context Protocol (MCP). - Experience deploying into sovereign, on-prem, or regulated environments (RBI/IRDAI/HIPAA-adjacent). - Early-stage or zero-to-one product experience; comfort with ambiguity and rapid iteration. What Success Looks Like (First 12 Months) - Aiva v1 architecture is documented, reviewed, and in production behind at least two accelerators (Convogent and Velogent) at one or more enterprisecustomers. - Hybrid retrieval meets agreed latency SLOs for voice workloads, with recall-quality benchmarks tracked release over release. - Multi-tenant isolation and governance controls pass enterprise security review at a regulated customer. - A 3–5 person Aiva team is hired, productive, and operating with a clear technical roadmap.