Multi-Agent Orchestration, LLM Infrastructure & Autonomous Intelligence

At Algolight, we live and breathe the future of artificial intelligence and the physical world.

Our mission it two-folder:

On the civilian side, we build labeled 3D information layers from all types of sensors—for smart cities, drones, autonomous vehicles, infrastructure, public safety, and far beyond.
On the defense side, we bring true real-time intelligence to the edge—anywhere, for any sensor, at any point on the map—enabling smart, real-time decisions in the field.

Border Team Mission

At Algolight, we are building the next generation of autonomous, multi-sensor AI systems designed to operate in real, operational environments — not just in theory. Our work spans from raw sensor data to high-confidence triage, reasoning, and autonomous decision-making, enabling scalable, robust, and intelligent systems that reduce cognitive load on operators and help drive safer, more effective national operations.

As part of a national effort to modernize border operations, you will be a core technical force — helping define, design, implement, and scale AI-driven architectures that tie together data, models, agents, and operational systems.

The Border team is evolving from:

sensor pipelines → model inference → human validation

into:

multi-agent systems → automated orchestration → adaptive learning loops → scalable autonomy

We aim to:

• Agentize scene understanding
• Agentize labeling and data curation
• Agentize model evaluation and retraining
• Agentize deployment and monitoring
• Agentize reasoning and triage
• Agentize system-level coordination

As a Principal Agentic AI Systems Engineer, you will architect and implement the agentic backbone of the entire border platform.

What You’ll Be Responsible For Some Of The Following

Multi-Agent System Architecture

Design and implement multi-agent architectures where agents:

• reason
• retrieve
• plan
• call tools
• delegate
• evaluate
• Define agent-to-agent contracts and context passing mechanisms.

Build orchestration layers that coordinate:

• perception agents
• context agents
• labeling agents
• evaluation agents
• retraining agents
• triage/decision agents

This includes:

• task planning
• execution graphs
• memory hierarchies
• conflict resolution
• deterministic overrides
• Agent Orchestration Infrastructure

Hands-on experience with (or equivalent depth understanding of):

• LangGraph
• CrewAI
• AutoGen
• OpenClaw-like tool-driven orchestration systems
• Custom orchestration frameworks

You will:

• Design execution graphs (DAGs + dynamic planning)
• Implement tool schemas and strict contracts
• Handle retries, fallbacks, validation layers
• Enforce deterministic constraints in critical flows
• Local & On-Prem LLM Systems

We operate in constrained environments.

You will do some of the following:

• Work with open-source models (LLaMA, Mixtral, Qwen, Mistral, etc.)
• Deploy models locally (vLLM, TensorRT-LLM, TGI, Ollama)
• Optimize quantization (GGUF, GPTQ, AWQ)
• Manage GPU utilization and batching strategies
• Implement inference routing strategies across models

You should understand:

• context window tradeoffs
• memory pressure
• streaming token generation
• latency vs reasoning depth

RAG, Memory & Knowledge Systems

Design and implement:

Multi-layer memory systems:

• short-term context memory
• long-term event memory
• vector-based retrieval memory
• structured knowledge graph memory
• Hybrid retrieval:
• vector search
• keyword search
• temporal filters
• spatial filters
• Grounded RAG pipelines over:
• sensor events
• historical logs
• operator feedback
• training artifacts

Tools may include:

• FAISS / Milvus / Weaviate
• OpenSearch / Elasticsearch
• Neo4j / Graph-based stores

Agentized Training & Labeling Automation

We want to automate:

• Human labeling workflows
• Data validation
• Dataset QA
• Edge-case detection
• Auto-curation of difficult samples
• Online learning triggers

You will design agents that:

• Review model outputs
• Flag inconsistencies
• Trigger relabeling
• Score data quality
• Decide when retraining is required

This is not academic.
This must work at scale.

Online Learning & Continuous Adaptation

Design pipelines that close the loop between:
inference → feedback → labeling → retraining → redeployment

Implement evaluation agents:

• regression detection
• drift monitoring
• performance decay tracking
• Coordinate retraining triggers under constraints.

Safety, Guardrails & Determinism

Build guardrail systems:

• structured output validation
• rule-based override layers
• confidence scoring
• Prevent hallucination in critical flows.
• Ensure explainability and traceability of agent decisions.

What We Are Looking For

Required

• 6+ years building production AI or distributed systems
• Real hands-on experience with agentic systems (not just notebooks)
• Deep understanding of RAG and memory architectures
• Experience deploying local models

Strong Advantages

• Experience in autonomy, defense, robotics, or ISR systems
• Knowledge graph integration
• Multi-modal reasoning (vision-language models)
• Experience automating training pipelines
• Understanding of perception pipelines

What Awaits You at Algolight

• Ownership of the agentic intelligence backbone of a national-scale system
• Freedom to design multi-agent systems correctly — not just hack demos
• Real operational constraints
• Deep collaboration with perception, radar, autonomy, and product teams
• A culture that values depth over hype