Python AI/ML with Full stack

Programmers.io; Dallas (TX); On-site; USD 180,000 - 230,000; Full time

Job summary

Programmers.io in Dallas is seeking an experienced AI Platform Engineer to build and scale a production multi-agent AI platform used by thousands of internal users across multiple business units. You’ll own end-to-end delivery from routing intents to tool governance.

You’ll work across LLM orchestration, agent selection, a fast FastAPI gateway, and a data API layer, ensuring reliability, latency, and cost controls.

Responsibilities

• LLM‑driven orchestrator to route user intent across agents and services
• Agent selection layer with hybrid retrieval and constrained JSON outputs
• Multi‑agent SDK / gateway using FastAPI with per‑agent tool registries
• Tool‑driven agents with dynamic tool composition and guardrails
• Data API layer ensuring LLMs never touch databases directly
• Partner‑team onboarding with versioned contracts and auto re‑embedding

Job description

Build and scale a production multi-agent AI platform serving thousands of internal users across multiple business units. Monthly release cadence, real users, real latency, real cost.

What You'll Own

• LLM-driven orchestrator that routes user intent across a portfolio of specialized agents — delegation, memory, response validation, capability discovery.
• Agent selection layer — hybrid retrieval (vector RAG over a capability registry) plus closed-set LLM selection with JSON-schema-constrained outputs.
• Multi-agent SDK / gateway — FastAPI service hosting many agents behind path-prefix routing, per-agent tool registries, session-scoped conversational context.
• Tool-driven agents — 15–30 tools per agent composed dynamically by an LLM; owns tool contracts, guardrails, and evaluation.
• Data API layer — parameterized endpoints between agents and databases; LLMs never touch DBs directly.
• Partner-team onboarding — versioned A2A contract, bring-your-own-agent registration, auto re-embedding.

Core AI Engineering

• Production LLM systems: RAG, tool/function-calling loops, structured outputs, hallucination guards, closed-set selection.
• Multi-agent orchestration: A2A protocols, session affinity, human-in-the-loop gating, kill switches, graceful degradation.
• Vector search + embeddings at scale (sub‑second retrieval over thousands of docs).
• Evaluation & safety: PII/PHI masking, audit trails, feedback-loop instrumentation, offline + online eval.

Platform / Infrastructure

• Python 3.11+, FastAPI, async I/O, Pydantic.
• Modern LLM stacks (Gemini, GPT, Claude) and agent frameworks (LangGraph, Agent SDKs).
• Cloud (GCP or AWS): Kubernetes, object storage, workflow orchestration, Vertex/Bedrock-class services.
• Observability: Prometheus, structured JSON logs, per-decision audit trails, p95 latency SLOs in seconds.

Ways of Working — Fast Turnaround, Ship-Fast

• Comfortable with short cycle times: spec → design → merged → deployed in days, not sprints. Monthly releases are the floor, not the ceiling.
• Bias to ship the smallest correct thing, verify in production, iterate. No polish before proof.
• Fluent with AI-assisted developer tooling (Claude Code, Cursor, agentic IDEs); reads and writes code with an LLM in the loop as a force multiplier.

Skill Curation & Reuse — Agentic Development Discipline

• Uses and extends the team's agentic SDLC skill library — capability intake, spec authoring, design docs, implementation plans, release‑impact artifacts, deployment records.
• Curates new skills when a workflow repeats: codifies patterns (accessibility, security/STRIDE, CI/CD, data‑source adapters, renderer standards) into reusable skills the whole team can invoke.
• Treats skills, prompts, and evals as first‑class artifacts — versioned, reviewed, and improved like code.
• Knows when to reach for a skill vs. write ad‑hoc: standard flows for standard work, creative bandwidth saved for novel problems.

You’ll Thrive Here If

• You've shipped LLM agents in production (not demos) with real users, latency, and cost constraints.
• You reason about routing, tool selection, and context strategy as first‑class design surfaces — not just prompt tuning.
• You own both the model layer and the platform underneath it (queues, auth, secrets, deploy, K8s).
• You move fast without breaking discipline: verification before completion, evidence before assertions.
• Regulated‑domain experience (healthcare, financial services) is a plus.

Bonus

• Contributions to agent frameworks, evaluation harnesses, or open A2A protocols.
• Patent / IP work in agentic systems, RAG, or multi‑agent orchestration.
• Prior experience authoring internal skill libraries, agent playbooks, or SDLC automation for AI teams.
• Strong hands‑on experience in designing, developing, and implementing Agentic AI solutions and frameworks at enterprise scale.
• Proven track record of delivering Agentic AI use cases in production environments, demonstrating measurable business value and outcomes.
• Ability to work independently with a high degree of ownership, accountability, and self‑motivation.
• Experience leading or contributing to large‑scale digital transformation initiatives for Fortune 500 and enterprise clients.
• Strong communication skills with the ability to clearly articulate solution approaches, business use cases, technical contributions, and delivered impact.
• Deep understanding of AI solution architecture, with the capability to confidently explain design decisions, technology choices, challenges, mitigations, and results.
• Demonstrated capability to conceptualize, architect, and build AI‑driven solutions end‑to‑end, from ideation through deployment and adoption.
• Ability to collaborate effectively with business and technology stakeholders while driving innovation and delivering tangible business outcomes.