Ai Platform Engineer (Agentic)
Confirmed on the employer's own hiring board on Sep 29, 2026. First seen by Alion on May 22, 2026. Salt scores C on the Alion truth index.
Builds and operates the core agentic AI platform that powers the AI SOC, AI Pen testing and AI Secure Code Review agents.
Owns the backend: agent orchestration framework, model integration layer, tool/function calling, memory and knowledge stores, evaluation harness, guardrails, and secure deployment on private cloud. Key Responsibilities:
* Design and implement the multi-agent orchestration framework using frameworks such as LangGraph, CrewAI, AutoGen, or custom Python.* Build the model-serving and routing layer supporting multiple LLM backends (self-hosted open-weights models and approved API providers), with caching, fallback and cost tracking.* Implement tool integrations: SIEM APIs, Tenable, ExtraHop, GitLab, Fortify, sandbox detonation, shell/code execution sandboxes, MITRE ATT&CK/D3FEND knowledge bases.* Build vector stores, RAG pipelines and long-term memory for agents; maintain curated security knowledge corpora.* Engineer the evaluation harness: golden datasets, regression tests, red-team prompts, quality and safety metrics per agent.* Implement guardrails and AI security controls per PLOT4AI, OWASP LLM Top 10, MITRE ATLAS and NIST AI RMF: prompt-injection defence, output filtering, tool-use authorization, data exfiltration controls, model supply-chain verification.* Package and deploy the platform on private cloud with full observability (traces, token usage, cost, latency, safety events).* Partner with the Principal AI Security Architect on threat modelling of the platform itself.
Goals
* Deliver a production-grade agentic AI platform that the three agent lines (SOC, Pentest, Code Review) can build on without reimplementing common primitives.* Ensure every deployed agent runs under enforced guardrails, authorization boundaries, and full audit logging.* Achieve reproducible agent quality through a rigorous evaluation and regression framework - no silent regressions between model or prompt updates.* Keep infrastructure and model costs transparent and under budget through routing, caching and quota controls.
Specific Objectives (SMART)
* Within 30 days: publish architecture, select orchestration framework, stand up dev environment, and deliver a ‘hello agent’ end-to-end trace.* Within 60 days: deliver v0.1 of the platform with tool-use, RAG, guardrails and evaluation harness; onboard the AI SOC agent as the first tenant.* Within 90 days: deliver v0.2 supporting the AI Pentest and AI Secure Code Review agents; publish threat model and AI RMF mapping.* Within 6 months: production-grade v1.0 with SLOs, on-call runbooks, and red-team validation.
Timeline & Engagement Model* 12-month contract.* Design + MVP: Months 1-3.* Multi-agent support: Months 3-6.* Production hardening: Months 6-12.
Required Skills & Experience:
* 5+ years backend engineering in Python (FastAPI, async, typing); strong systems design.* Hands-on experience building LLM agent systems with LangGraph / LangChain / CrewAI / AutoGen or equivalent.* Production experience with vector DBs (pgvector, Qdrant, Weaviate), embedding pipelines, and RAG.* Strong MLOps / LLMOps: model serving (vLLM, TGI, Ollama), evaluation frameworks, tracing (OpenTelemetry, LangSmith, Langfuse).* Cloud-native delivery: Docker, Kubernetes, Helm, CI/CD via GitLab.* Working knowledge of OWASP LLM Top 10, NIST AI RMF, MITRE ATLAS, PLOT4AI.* Security mindset: secrets handling via Vault, least-privilege tool access, auditability.
Salt is acting as an Employment Business in relation to this vacancy.