关于项目 / About the Project
Oakcare 是新加坡养老护理平台。现有代码库基于 Python + Google ADK + Gemini 2.5 构建，包含 WhatsApp 接待机器人、护理人员简历处理管道和自动评分/匹配引擎。代码库需要稳定化和加固，然后正式上线并扩展到更多护理垂直领域。本岗位主导 AI/LLM 管道工作，同时参与后端和基础设施的共建。
Oakcare is a Singapore-based eldercare placement platform. The existing codebase is built on Python + Google ADK + Gemini 2.5, including a WhatsApp intake concierge, caregiver profile enrichment pipeline, and automated scoring/matching engine. The codebase needs stabilisation and hardening before we officially launch and scale to additional care verticals. This role leads AI/LLM pipeline work and contributes to shared backend and infrastructure tasks.
当前重点 / Current Focus: Stabilisation
-
修复 AI 管道的静默失败、竞态条件和数据丢失问题
-
为所有 LLM 响应添加结构化输出验证（JSON Schema）
-
实现重试逻辑、错误处理、回退搜索策略
-
修复评分确定性 — 固定规则改用确定性函数，不用 LLM
-
添加 LLM 可观测性和测试框架（golden file 测试）
-
修复数据新鲜度 — 存储出生日期而非固定年龄，查询时动态计算
-
参与 CI/CD 搭建、PostgreSQL 维护、结构化日志、密钥修复
-
Fix silent failures, race conditions, and data loss in the AI pipeline
-
Add structured output validation (JSON Schema) for all LLM responses
-
Implement retry logic, error handling, and fallback search strategies
-
Fix scoring determinism — replace fixed rubrics with deterministic functions
-
Add LLM observability and test framework (golden file tests)
-
Fix stale data — store DOB instead of fixed age, recalculate at query time
-
Contribute to CI/CD setup, PostgreSQL maintenance, structured logging, secrets remediation
后续阶段 / Next Phase (After Stabilisation)
稳定化完成后，项目将进入基础设施迁移和多垂直领域扩展，可能涉及：LLM 网关搭建（多模型提供商路由）、提示词和评分配置外部化、对话审计日志恢复。具体范围在第一阶段结束时确认。
Following stabilisation, the project moves into infrastructure migration and multi-vertical expansion, which may include: LLM gateway for multi-provider routing, externalising prompts and scoring config, and re-enabling the conversation audit trail. Scope to be confirmed at the end of Phase 1.
技术要求 / Requirements
-
精通 Python（async/await、状态管理、测试）
-
生产级 LLM 集成（Gemini、OpenAI 或类似）
-
结构化输出 / JSON Schema 强制约束
-
Agent 框架经验（Google ADK、LangChain 或类似）
-
PostgreSQL 和 CI/CD 基础
-
加分：Go、K8s 基础、阿里云 / 百炼经验、Twilio
-
Strong Python — async/await, state management, testing
-
Production LLM integration — Gemini, OpenAI, or similar
-
Structured output / JSON schema enforcement
-
Agent framework experience — Google ADK, LangChain, or similar
-
PostgreSQL and CI/CD fundamentals
-
Bonus: Go, K8s basics, Alibaba Cloud / Model Studio, Twilio
技术栈 / Tech Stack
当前 / Current:
Python · Google ADK · Gemini 2.5 · PostgreSQL (via Xano) · Redis · Twilio · Brevo · Stripe · Firecrawl · React
第二阶段可能新增 / Potential Phase 2 additions:
阿里云百炼 · LangSmith / Langfuse · Go API 网关 · K3s / GKE · ArgoCD
薪资 / Compensation
面议，项目制。申请时请附上 LLM 管道或数据管道相关工作示例。
Negotiable, project-based. Please include examples of LLM pipeline or data pipeline work with your application.
8
可以做的可联系
你这个现在应该有技术合伙人？