C2C
AI/ML Engineer – Technical Skill Set (Agentic AI Focus)
Location: Remote
- Core Programming & Systems Skills
- Python (expert level) for ML, orchestration, and agent logic
- Strong understanding of async programming, concurrency, and task scheduling
- Foundations of Agentic AI
- Design and implementation of autonomous AI agents capable of:
- Multi-step reasoning and planning
- Goal decomposition and task orchestration
- Dynamic decision-making under uncertainty
- Experience with agent architectures:
- React, Plan-and-Execute, Reflexive agents
- Hierarchical / multi-agent systems
- Tool-augmented and function-calling agents
- Understanding of stateful vs stateless agents and memory management
- Large Language Models (LLMs)
- Hands-on experience with LLMs (OpenAI, Azure OpenAI, Anthropic, open-source models)
- Prompt-engineering techniques for:
- Reasoning (Chain-of-Thought, Self-Reflection)
- Planning and critique loops
- Instruction following and tool use
- Experience with:
- Few-shot and zero-shot prompting
- Model selection trade-offs (latency, cost, context length)
- Knowledge of fine-tuning / adapters (LoRA) is a plus
- Agent Frameworks & Tooling
- Practical experience with agent frameworks, such as:
- LangGraph / LangChain (agents, tools, memory)
- Semantic Kernel
- AutoGen, CrewAI, or similar
- Ability to build custom agent orchestration layers beyond frameworks
- Tool abstraction and execution safety (timeouts, retries, sandboxing)
- Memory, Context & Knowledge Augmentation
- Design of agent memory systems:
- Short-term (conversation/state memory)
- Long-term (episodic, semantic memory)
- Retrieval-Augmented Generation (RAG):
- Vector databases (FAISS, Pinecone, Azure AI Search, etc.)
- Embedding selection and chunking strategies
- Techniques for context management and compression
- Knowledge graph-augmented or hybrid memory (plus)
- Planning, Reasoning & Control
- Experience implementing:
- Task planners (step planning, re-planning)
- Constraint-based execution
- Feedback and self-correction loops
- Understanding of:
- Tool reliability scoring
- Guardrails and action validation
- Failure detection and graceful recovery