Agentic AI Engineer (AI)

TL;DR

Agentic AI Engineer (AI): Build agentic systems for silicon design by training and adapting LLMs/SLMs, engineering design-context retrieval (RAG, prompt scaffolds, tool-calling), and grounding agents with tuned knowledge graphs and vector/graph databases. Focus on optimizing accuracy/latency/cost, hardening production agent tools and guardrails, and preventing regressions through offline benchmarks and online telemetry.

Location: CARY

Company

hirify.global builds applied AI systems for silicon design.

What you will do

• Develop models: train, fine-tune, distill, and evaluate LLMs/SLMs and embedding models for EDA-specific tasks (LoRA/PEFT, instruction tuning, DPO/GRPO, eval harnesses).
• Engineer design context: build retrieval pipelines, prompt scaffolds, and tool-calling specs to supply RTL/scripts/logs/reports within the right token budget.
• Tune knowledge and databases: design schemas and optimize ingestion/queries for graph DBs (Neo4j, ArangoDB, NebulaGraph) and vector stores (Qdrant, Weaviate, pgvector, Chroma).
• Build agent building blocks: implement and harden agent tools, memory, multi-hop reasoning patterns, and guardrails; triage and iterate on production failures.
• Own data pipelines: curate, clean, and label datasets from EDA artifacts; build synthetic-data and self-improvement loops where appropriate.
• Measure quality: create offline benchmarks and online metrics, define what “good” means for chip-design agents, and keep regressions out of production.

Requirements

• BS/MS/PhD in CS, EE, ECE, AI/ML, or a closely related field (graduating in 2025–2026; recent grads welcome).
• Strong deep learning and transformer/LLM fundamentals (attention, tokenization, context windows, decoding).
• Hands-on experience with at least two of: LLM fine-tuning, RAG/retrieval, agentic frameworks, knowledge graphs, vector databases.
• Solid Python engineering with comfort in PyTorch and Hugging Face; writes clean, tested, version-controlled code.
• Strong written and verbal communication; bias to ship working code.

Nice to have

• Internship in AI/ML at a product company or research lab with shipped artifacts.
• Hands-on with agentic frameworks (LangGraph, AutoGen, Cursor SDK, Claude Code, MCP-based tool-calling stacks).
• Experience with graph DBs and/or vector DBs (Neo4j, ArangoDB, NebulaGraph, Qdrant, Weaviate, pgvector, Chroma, Milvus).
• ML systems/infra exposure (vLLM, TGI, Triton, distributed training, GPU performance tuning, quantization).
• Coursework/projects in compilers, formal methods, HDLs (Verilog/SystemVerilog/Chisel), or EDA tools.

Culture & Benefits

• Work with senior AI engineers and chip-design domain experts on core pillars of an agentic AI stack.
• From day one, write production code that ships into customer-facing AI products.
• Pairing and learning support to ramp up on the EDA flow.
• Emphasis on rigorous evaluation and keeping regressions out of main.