דרושים » AI » מפתח /ת Full Stack מנוסה - Node.js - React - TypeScript - AI Development

Join team to build and extend the AI assistant embedded across our product suite. The platform is a production LLM agent system that's integrated into multiple host applications, backed by a RAG knowledge base and an evaluation-driven development workflow.

You'll work end-to-end: agent design, tool implementation, retrieval quality, integration into existing product UIs, cloud infrastructure, and evaluation/observability. The platform already ships to customers - you'll extend it, raise its quality bar, and help define where it goes next.

What you'll do

Design and evolve agents - build LLM agents with tool use, routing, and human-in-the-loop flows.
Implement tools and integrations - expose product capabilities to the agent, with multi-tenant context, via internal APIs and MCP servers.
Own retrieval quality - contribute to our RAG pipeline end-to-end: ingestion, embeddings, vector search, and reranking.
Define and evolve host integration contracts - collaborate with host application teams to integrate the assistant into product UIs built on different frontend stacks. You own the shared remote module and the integration API; host teams own their stacks.
Drive evaluation-led development - write evaluators (rule-based, LLM-as-judge, multi-turn), maintain CI eval gates, and use traces and feedback to debug production behavior.
Operate the platform - own deployments, observability, and the performance and cost of LLM-backed services.
Establish engineering practices** for AI-specific work: prompt versioning, eval coverage, testing, and code review.

We are looking for a hands-on wireless systems engineer to lead the development of a multi-RAT digital twin for our Open RAN solution. The digital twin will execute production RAN software-beginning with scheduler and MAC behavior-in a closed loop with PHY, channel, UE, traffic, and network models. It will allow engineering teams to design, evaluate, and compare features for LTE, 5G NR, and 2G without requiring a dedicated physical radio setup for every development cycle.
This is a senior individual-contributor role at the intersection of wireless systems, simulation, production software, and AI/ML. You will evolve an existing LTE end-to-end simulator into a scalable engineering platform for feature development, regression testing, performance optimization, and evidence-based pre-validation. Initial use cases include MAC scheduler and link-adaptation improvements, power control, mobility and interference scenarios, and neural-network-assisted channel estimation.
The successful candidate will understand that a useful digital twin must be both fast and trustworthy. You will define multiple fidelity levels-from rapid surrogate models to full PHY processing-and establish repeatable methods for calibrating the twin against lab or field reference data. The goal is to reduce dependence on continuous lab access while maintaining clear, measurable confidence in the simulation results.
What you will do:
Own the technical architecture and roadmap for a modular, multi-RAT RAN digital twin
covering LTE, 5G NR.
Integrate production MAC and scheduler software into deterministic, per-TTI/slot closed-loop simulations
through stable and maintainable interfaces.
Model the interaction among scheduler decisions, PHY processing, propagation channels, UE behavior, traffic, interference, mobility, HARQ, link adaptation, and power control.
Extend the current LTE simulation capability and define reusable abstractions that support additional 5G NR and
2G stacks without duplicating the platform.
Design a fidelity ladder that combines high-fidelity PHY execution with faster calibrated models or lookup/surrogate backends, selecting the least expensive model that is valid for each engineering question.
Develop and evaluate AI/ML-based RAN capabilities, including neural channel estimation, learned link adaptation or scheduling policies, and ML-based PHY or channel surrogates.
Build representative datasets and experiment pipelines; establish conventional algorithmic baselines;