Agentic AI Developer

We are seeking a hands-on Agentic AI Developer to design implement and productionize autonomous and tool using AI agents that run on premises or at the edge with no dependency on SaaS LLM APIs You will develop planners tool chains memory and multiagent collaboration build host local models and integrate agents safely with internal systems in secure air gapped or private environments This role blends LLM systems tool orchestration retrieval reasoning secure sandboxing and MLOps for local inference

Key Responsibilities

Agent Architecture Orchestration

• Design agentic architectures single and multiagent with planning reflection tool calling and long-term memory eg ReAct Reflexion style loops task decomposition self-critique
• Build tool use interfaces for enterprise systems REST Graph QL DBs message buses file systems shell tasks and domain specific tools with strict policy guards capability scoping and sandboxing
• Implement deterministic control flows graphs DAGs finite state machines for reliability support interpretability rollbacks retries and human in the loop checkpoints

Local Models Inference

• Stand up and optimize locally hosted models not SaaS( Llama Mistral Mixtral DeepSeek Phi families code models for tool synthesis ASRTTS if required)
• Use vLLM Triton llama.cpp Ollama for inference optimize with quantization GPTQ AWQ GGUF tensor parallelism KVcache batching leverage CUDAROCm for GPUs
• Implement RAG pipelines chunking hybrid search reranking with Qdrant,Milvus,FAISS apply meta data security filters and freshness lineage

Safety Governance Security

• Enforce policy constraints prompt hardening data redaction tool execution sandboxing name spaces sec comp SELinux/AppArmor and network egress control
• Build audit logs traces and activity lineage for agent actions and tool invocations surface explainability and decision rationales
• Handle PII PHI PCI safeguards RBACABAC secrets management license governance OSS commercial model licenses and offline updates

Reliability Evals Observability

• Define agent evals task success tool call precision recall chain of thought proxies if applicable hallucination risk RAG quality RAGAS latency and cost compute tokens
• Integrate telemetry Open Telemetry traces and LLM run logs build dashboards Grafana ELK and playbooks for incident response
• Create offline air gapped CICD container images SBOMs reproducible builds model registries and controlled rollouts

Collaboration Delivery

• Partner with Product Domain SMEs to translate workflows into agent tools and policies
• Author design docs risk assessments and user guides run threat modelling for agent integrations
• Mentor engineers in agent patterns prompt response schemas and safe tool wiring

Job Requirements

• Mandatory Skills : AI/GenAI Research
• Strong Agentic AI developer (not SaaS cloud based) with relevant experience of 5 years.