Trivita AI is hiring Middle/Senior Agentic AI Engineers with expertise in Python, LLMs, AI agents, RAG, evaluation, and production deployment.
Trivita AI develops AI-powered products and platforms for businesses, including AI assistants, workflow automation agents, knowledge search, and speech recognition. Our systems combine Python backends with commercial and self-hosted LLMs running on GPUs, deployed across Kubernetes, cloud platforms, and on-premise infrastructure.
As a Middle/Senior Agentic AI Engineer, you will design, develop, and operate end-to-end AI agent systems, covering agent architecture, tool integration, knowledge retrieval, backend services, security, quality evaluation, and production operations.
System success is measured through real-world metrics such as task completion rate, accuracy, latency, and cost per task.
Job description
AI Agent development
- Design and build AI agents capable of planning, tool calling, and executing multi-step workflows.
- Implement checkpointing, durable execution, and human-in-the-loop approval for high-risk actions.
- Develop tool integrations using function calling and Model Context Protocol (MCP), ensuring clear schemas, timeouts, and appropriate access controls.
- Manage agent context and memory, including conversation summarization, short-term and long-term memory, prompt caching, and token budgets.
- Integrate commercial LLMs from OpenAI, Anthropic, and Gemini, as well as self-hosted models through vLLM.
- Implement structured outputs, model routing, retries, fallbacks, rate limiting, and cost tracking.
Knowledge retrieval and backend engineering
- Build retrieval pipelines using chunking, indexing, hybrid search, reranking, query rewriting, and metadata filtering.
- Apply knowledge graphs and Graph RAG for use cases requiring structured relationships.
- Develop Python backends using FastAPI, asynchronous programming, APIs, data models, background jobs, and queues.
- Implement real-time streaming through SSE or WebSockets.
- Optimize retrieval quality, scalability, and system performance.
AI evaluation and security
- Develop AI evaluation systems using golden datasets, evaluation metrics, LLM-as-judge, and tracing.
- Establish automated evaluation workflows to detect quality regressions before releases.
- Analyze failure cases, identify root causes, and conduct A/B tests or ablation studies to validate improvements.
- Implement defenses against prompt injection, input/output validation, personal data protection, and audit logging.
- Collaborate with QA teams on labeling guidelines and golden datasets.
Production eeployment and operations
- Package and deploy AI services using Docker, Kubernetes, Helm, and GitOps.
- Integrate OpenTelemetry for system tracing and troubleshooting.
- Collaborate with DevOps teams on autoscaling and GPU resource management.
- Apply AI coding assistants to coding, testing, reviews, and debugging while validating outputs before production deployment.
- Research emerging techniques from scientific publications and adopt them when supported by measurable evaluation results.
Requirements
Experience and technical skills
- Middle: 3+ years of software engineering experience with Python as the primary language, including at least 1 year developing LLM-based systems used by real users.
- Senior: 5+ years of software engineering experience with Python as the primary language, including at least 2 years developing LLM-based systems used by real users.
- Experience deploying at least one AI agent or RAG system to production, with the ability to explain its architecture, failure modes, and evaluation metrics.
- Strong Python skills, including async/await, type hints, Pydantic, and pytest.
- Experience operating production backends with FastAPI, PostgreSQL, and Redis.
- Strong understanding of embeddings, hybrid search, reranking, and retrieval metrics such as Recall@k, MRR, and nDCG.
- Ability to distinguish retrieval errors from generation errors.
- Experience developing LLM evaluation systems using datasets, metrics, calibrated LLM-as-judge, and error analysis.
- Understanding of tokenization, context windows, sampling, hallucinations, prompt injection, and quality-latency-cost trade-offs.
- Proficiency in Docker, Kubernetes, Helm, and at least one major cloud platform (AWS, GCP, or Azure).
- Experience using AI coding assistants for development, testing, research, documentation, and workflow automation while complying with data security policies.
- Ability to read technical documentation and research papers in English.
Level-specific expectations
Middle Agentic AI Engineer
- Build and improve agent/RAG features within an existing system architecture.
- Develop evaluation suites and tests for assigned components.
- Use tracing to investigate failures and improve system quality.
Senior Agentic AI Engineer
- Design agent architecture, tool integrations, safety layers, and evaluation-gated release workflows.
- Make technical decisions balancing quality, latency, and cost.
- Provide technical leadership and mentor Middle engineers.
Technology stack
- Programming & Backend: Python, FastAPI, Pydantic, PostgreSQL, Redis.
- AI Agent & Integration: LangGraph, MCP, OpenAI, Anthropic, Gemini.
- Retrieval & Knowledge: Qdrant, Elasticsearch/OpenSearch, Neo4j.
- LLM Serving: vLLM.
- Observability: Langfuse, OpenTelemetry.
- Infrastructure: Docker, Kubernetes, Helm, ArgoCD, AWS, GCP, Azure.
First 6 months expectations
- Deploy at least one agent/RAG feature to production with an evaluation suite and tracing.
- Ensure all changes to prompts, models, or retrieval components pass automated evaluation before release.
- Deliver at least one measurable improvement in task completion rate, Recall@k, or cost per task, supported by documented results.
Preferred qualifications
- Production experience with knowledge graphs or Graph RAG using Neo4j or equivalent tools.
- Experience self-hosting LLMs or speech models on GPUs using vLLM, SGLang, or NVIDIA Triton.
- Experience building MCP servers or agent tool platforms.
- Familiarity with LangGraph, LlamaIndex, DSPy, Ragas, DeepEval, or Langfuse.
- Experience with Qdrant, pgvector, or Elasticsearch/OpenSearch.
- Knowledge of multi-agent systems, long-term agent memory, fine-tuning, or reinforcement learning for agents.
- Experience developing agents for healthcare, legal, financial, or sensitive-data environments.
- Research publications, open-source contributions, or public technical writing.
Contact information
- Address: No. 01, Street 104, Quarter 3, Binh Trung Ward, Ho Chi Minh City
- Phone: [protected info]