Staff Software Engineer, Faraday Future

• Bachelor’s degree or above in Computer Science, Software Engineering, or a related discipline
• 8–10+ years software engineering with 3+ years in LLM/agentic systems or production NLP/IR
• Strong backend skills (Python, Node.js or Java), APIs (REST/GraphQL/WebSocket), and cloud delivery (Docker/K8s on AWS or Google Cloud)
• Google ADK proficiency: multi-agent graphs, tool/function schemas, memory/sessioning, built-in evaluators; deployment on Vertex AI Agent Builder/Agent Engine
• Retrieval & vector search: embeddings, rerankers, vector DBs (pgvector/FAISS/Milvus), chunking/routing, and latency/cost-aware caching
• Evals & guardrails: factuality/grounding tests, safety filters, jailbreak resistance, and human-in-the-loop workflows
• Observability: tracing tokens/latency, cost controls, logs/metrics/traces; CI/CD and canarying for prompts/tools/policies; feature-flag frameworks
• Security foundations: secrets management/rotation, signed webhooks, OAuth2/OIDC, mTLS where appropriate

Desirable

• As a Staff Software Engineer (Agent Systems) in our Crypto Department, you will architect and implement agentic systems that convert user intent into safe, explainable, multi-step research and execution workflows
• You’ll design Google Agent Development Kit (ADK)–based multi-agent graphs, define tool/function schemas, integrate retrieval pipelines, and establish guardrails, evaluation, and observability to deliver grounded, low-latency experiences across web and mobile
• Agent Orchestration (Google ADK): Build single/multi-agent graphs (planner–tool–critic loops), manage state/memory/sessioning, and use ADK built-in evaluators for trajectory/factuality; deploy with Vertex AI Agent Builder/Agent Engine under clear latency/cost/grounding SLAs
• Retrieval & Grounding: Implement RAG that fuses curated sources with on-chain/market/social signals; source attribution/citations and freshness controls; reranking where needed
• Tools & Integrations: Integrate REST/GraphQL/WebSocket providers, safe browser/HTTP fetch, and exchange/explorer APIs with rate-limit/backoff/reconnect; define typed adapters and sandboxed tool scopes
• Realtime UX & BFF: Stream intermediate results and progressive answers; expose agent endpoints to web/mobile via BFF with typed contracts and backpressure handling
• Reliability & Safety: Enforce prompt-injection defenses, PII redaction, policy filters, allow/deny-lists, and safe execution/rollback; maintain gold sets and red-teaming suites
• Evals & Observability: Instrument tokens/latency/cost; build offline eval sets and online A/B/interleaving tests; track citation coverage and tool-action success; version prompts/policies and canary new tools
• Documentation & Standards: Author ADRs, tool specs, and runbooks; coordinate with Security/Compliance and Product