The posting
Parallel Wireless is a U.S.-based pioneer in Open RAN innovation, transforming how mobile networks are built, optimized, and powered. Through our GreenRAN™ portfolio, we help operators deliver secure, energy-efficient, automated, and flexible connectivity across 2G, 3G, 4G, 5G, and the path toward 6G. Our software-centric, hardware-agnostic approach brings intelligence into the RAN while helping customers reduce complexity and total cost of ownership.
Parallel Wireless is looking for a hands-on technical leader to build and operate a secure local large-language-model platform for the company. The platform will allow engineering and business teams to use generative AI with proprietary source code, product documentation, technical standards, test artifacts, support knowledge, and other approved internal data while keeping sensitive information within company-controlled environments.
This is a senior individual-contributor role spanning applied LLM engineering, platform architecture, search and data pipelines, security, and production operations. You will turn promising prototypes into a dependable internal capability: selecting and optimizing open-weight models, building permission-aware retrieval, creating reusable APIs and tools, integrating with existing engineering workflows, and establishing objective ways to measure quality, safety, latency, capacity, and business value.
The successful candidate will understand that a useful enterprise LLM is more than a model and a chat interface. It requires trustworthy source grounding, strong access controls, repeatable evaluation, careful tool permissions, observable production services, and an operating model that keeps data, indexes, prompts, models, and dependencies current. You will make pragmatic build-versus-buy decisions and choose the simplest approach—search, retrieval-augmented generation (RAG), prompting, workflow automation, or model adaptation—that meets each use case.
Initial use cases may include engineering knowledge discovery, source-code understanding, troubleshooting assistance, technical-document Q&A and summarization, test and log analysis, and drafting structured engineering artifacts. The platform should be extensible to additional approved use cases as needs and model capabilities evolve.
What you will do:
- Own the architecture and technical roadmap for a secure, reliable, and maintainable local LLM platform deployed in Parallel Wireless-controlled infrastructure.
- Partner with engineering, product, support, IT, information security, legal, and domain experts to prioritize high-value use cases and translate them into measurable product and platform requirements.
- Build a modular inference and model-gateway layer with stable APIs, model routing, streaming, concurrency controls, quotas, and the ability to change models or serving backends without rewriting every application.
- Evaluate open-weight language, code, embedding, reranking, and, where useful, multimodal models against PW-specific tasks; document model provenance, licenses, limitations, security posture, hardware needs, and total cost of ownership.
- Optimize serving across available CPU, GPU, and accelerator resources using techniques such as continuous batching, caching, parallelism, quantization, and right-sized context limits while protecting output quality.
- Design and operate RAG and enterprise-search pipelines for approved repositories, wikis, tickets, standards, design documents, test results, logs, and support content, including parsing, chunking, metadata, embeddings, hybrid retrieval, reranking, freshness, citations, and deletion.
- Enforce source-system permissions throughout ingestion and retrieval so that the platform never exposes content a user is not authorized to access; integrate with company identity, SSO, role-based access control, secrets management, and audit logging.
- Establish versioned evaluation datasets and automated offline and online evaluation for retrieval quality, groundedness, factual accuracy, citation quality, code correctness, task completion, latency, safety, and refusal behavior.
- Create release gates and reproducible regression tests for changes to models, prompts, tools, embeddings, retrieval logic, indexes, and serving configurations; support canary releases, rollback, and clear approval paths.
- Implement end-to-end observability for model and agent workflows, including traces, errors, time to first token, inter-token latency, throughput, queue time, resource utilization, saturation, availability, and user feedback.
- Design safe tool-calling and agent workflows with least-privilege access, sandboxing, input and output validation, bounded execution, human approval for consequential actions, and complete traceability.
- Integrate the platform into the tools employees already use—such as developer environments, source-control and CI workflows, knowledge systems, ticketing systems, and internal applications—through reusable SDKs, APIs, and reference implementations.
- Build the operational foundations for production use: CI/CD, configuration and model registries, backups, disaster recovery, capacity planning, dependency and vulnerability management, incident response, and lifecycle policies for models and data.
- Protect proprietary and personal information through network isolation, encryption, retention controls, redaction where appropriate, secure logging, and defenses against prompt injection, data poisoning, unsafe output handling, and model-supply-chain risks.
- Determine when prompt or retrieval improvements are sufficient and when parameter-efficient fine-tuning, distillation, or other adaptation is justified by measured quality gains.
- Make the platform usable beyond the core AI team through documentation, examples, training, office hours, and hands-on collaboration; use telemetry and structured feedback to improve adoption and effectiveness.
- Communicate architecture decisions, quality evidence, risk, capacity, and roadmap tradeoffs clearly to technical and business stakeholders.
What you bring:
- BSc or MSc in Computer Science, Computer Engineering, Electrical Engineering, Data Science, or a related field, or equivalent practical experience.
- Typically 7+ years of hands-on experience in production software, ML platform, search, data, or infrastructure engineering, including meaningful recent experience shipping LLM-powered systems; exceptional candidates with equivalent depth are welcome.
- Strong Python engineering skills and experience designing maintainable APIs, services, libraries, and data pipelines. Experience with Go, Java, or C/C++ is an advantage.
- Strong understanding of transformer-based language models and production inference, including tokenization, context management, batching, KV caching, parallelism, quantization, structured output, tool calling, and common model failure modes.
- Demonstrated experience building production RAG or enterprise-search systems using embeddings, vector and/or lexical search, metadata filtering, reranking, source attribution, and systematic retrieval evaluation.
- Experience defining task-specific LLM evaluations using representative datasets, strong baselines, domain-expert review, automated metrics, human feedback, error analysis, and regression thresholds.
- Experience deploying and operating containerized services on Linux using Docker and Kubernetes or an equivalent orchestration environment.
- Practical experience with GPU-backed model serving, performance profiling, capacity planning, monitoring, and reliability engineering.
- Strong knowledge of distributed-system fundamentals, authentication and authorization, API security, secrets handling, encryption, auditability, and data lifecycle controls.
- Experience with Git, automated testing, CI/CD, infrastructure as code, observability, and production incident response.
- Sound technical judgment about quality, security, maintainability, hardware efficiency, and total cost—not just model benchmark scores.
- Ability to lead an ambiguous, cross-functional initiative, explain complex AI behavior in plain language, and help other teams ship safely on a shared platform.
Nice to have:
- Experience operating LLMs in on-premises, private-cloud, restricted-network, or air-gapped environments.
- Hands-on experience with current inference runtimes and serving systems such as vLLM, SGLang, TensorRT-LLM, llama.cpp, Ray Serve, KServe, Triton, or equivalent technologies.
- Experience optimizing inference on NVIDIA and/or AMD GPUs using CUDA, ROCm, profiling tools, tensor parallelism, pipeline parallelism, speculative decoding, prefix/KV caching, or related techniques.
- Experience with model and experiment registries, LLM tracing and evaluation platforms, vector databases, hybrid-search engines, and production data-orchestration frameworks.
- Experience with parameter-efficient fine-tuning methods such as LoRA/QLoRA, dataset curation, synthetic-data generation, distillation, and post-training evaluation.
- Experience building code intelligence, repository-aware assistants, developer tools, or IDE and CI integrations for large C/C++ and Python codebases.
- Familiarity with Active Directory or another enterprise identity provider, fine-grained document authorization, data-loss prevention, secure software supply chains, model licensing, and AI governance.
- Experience red-teaming LLM or agent systems for prompt injection, sensitive-data disclosure, poisoned retrieval content, excessive agency, and insecure output handling.
- Knowledge of telecommunications, 3GPP, RAN/Open RAN, cloud-native network functions, or technical-support workflows.
- Experience working across heterogeneous compute platforms and making performance, energy, and TCO tradeoffs for enterprise AI workloads.
- Contributions to relevant open-source AI, search, MLOps, or infrastructure projects.
