Lead AI Engineer, Applied AI at Trimble Transportation
Pawan Kumar
I turn applied AI ideas into systems teams can ship and run.
10+ years in software·5+ in ML/AI·Technical guidance for 20+ engineers
About
I’m a Lead AI Engineer in Trimble Transportation’s Applied AI group in Germany. I guide architecture, technical direction, and shared engineering standards for applied AI systems, while mentoring a group of 20+ engineers. My focus is getting useful AI systems into production and keeping them dependable.
I got into tech in 2015 as a software engineer at Infosys, building enterprise applications. Then I moved to Germany for my Master's and fell deep into the ML/AI rabbit hole. Since then I've worked across research and industry: reinforcement learning for Kubernetes scheduling at Leibniz University, production ML at DISH Digital where my approach to model and infrastructure design cut costs by 30%, and now Applied AI engineering at Trimble. Alongside DISH I co-founded Arelis AI as a part-time, self-employed venture, serving as technical co-founder and Head of AI and owning the early product and architecture for GenAI governance and compliance (ISO 42001, EU AI Act readiness).
What I enjoy most is the messy middle: turning an unclear problem into a sound architecture, deciding where an LLM helps, and designing for evaluation, observability, and reliability. I work across teams to set technical direction, mentor engineers, and build shared standards that help applied AI move from prototype to production.
At a glance
How I work
The principles behind my architecture and technical guidance.
Production, not demos
Start by deciding which parts of a problem actually need an LLM. Then build for real operating conditions: observability, fallbacks and on-call from day one.
Governance by design
ISO 42001 and EU AI Act readiness are design inputs, not paperwork. Auditability and responsible-AI practices belong in the platform, not bolted on at the end.
Standards that compound
Reference architectures and shared standards let product teams ship reusable, governed AI capabilities instead of one-off prototypes.
Teams that grow
Mentoring and code reviews are core work. I help engineers grow across the stack and turn business problems into technical roadmaps they can own.
Experience
- 20+ engineers technically guided
- ~45% lower LLM cost
- Up to 60% lower latency
- Set the technical direction for Applied AI across internal and customer-facing products, establishing architecture guidance and shared standards for a group of 20+ engineers.
- As technical lead for Arc Agent, guided its architecture from product goals through global launch, with production readiness and governance built into the design.
- Designed multi-model routing and prompt caching across Anthropic, OpenAI, and Gemini, reducing LLM cost per task by ~45% and response latency by up to 60%.
- Established shared standards for AI-assisted software design, implementation, review, and verification; mentor junior and mid-level engineers and spoke at Trimble AI Summit 2026.
- Built cost and adoption analytics for leadership, connecting AI spend and usage to pricing and go-to-market decisions.
- ISO 42001 & EU AI Act
- Web Summit Lisbon 2024
- Co-founded an early-stage GenAI startup focused on AI governance and compliance, owning product vision, architecture, and end-to-end development as technical founder.
- Recruited and mentored the founding engineering contributors, setting technical direction during the company's formative stage.
- Shaped the platform around ISO 42001, EU AI Act readiness, auditability, and responsible AI adoption.
- Selected model and tooling vendors on cost, compliance, and capability under startup budget constraints.
- Exhibited Arelis AI at Web Summit Lisbon 2024, driving product positioning, networking, and ecosystem engagement.
- 30% cloud cost savings
- 40% less manual workload
- 1M+ synthetic samples
- 5+ business units
- Directed the technical strategy for 5+ end-to-end AI initiatives, including agentic orchestration and multimodal LLM/ML pipelines; GenAI programmes cut manual processing and annotation workload by 40%.
- Built a multilingual synthetic data engine generating 1M+ samples, raising model accuracy in underrepresented and low-resource language domains.
- Standardized MLOps across 5+ business units (enterprise-grade CI/CD, observability, governance), shortening model iteration cycles by over 50%.
- Delivered a secure, scalable AI deployment framework on GCP (Terraform, Kubeflow, Airflow) with 99.9% uptime and 30% cloud cost savings, owning capacity planning and cloud spend for the ML estate.
- Established AI governance frameworks across AI use cases, integrating responsible AI practices.
- 40% better resource utilization
- EU-funded BRAINE project
- Designed RL-based scheduling algorithm for Kubernetes clusters under EU-funded Braine Project, improving resource utilization by 40%.
- Researched advanced RL algorithms, developing and maintaining distributed ML experiments on Kubernetes clusters for real-time applications.
- 99.9% system availability
- Managed deployment and runtime operations for enterprise Java applications on WebSphere and WebLogic, ensuring 99.9% system availability.
- Collaborated with cross-functional teams to troubleshoot deployment pipelines and support production rollouts.
Skills
Applied AI architecture, production GenAI, evaluation and governance, supported by cloud and MLOps experience.
OpenAI
Anthropic
Gemini
Llama
Mistral
Hugging Face
Ollama
Groq
LangChain
LlamaIndex
CrewAI
MCP
n8n
Claude Code
Cursor
Hermes Agent- Deep Agents Planning, sub-agents and long-running agent execution
- Open Knowledge Format Agent-maintained Markdown knowledge bases (LLM Wiki)
- A2A protocol Interoperable agent-to-agent communication
AWS
Bedrock
GCP
Vertex AI
Azure
Azure AI
Docker
Kubernetes
Terraform
Python
Databricks
MLflow
dbt
Pinecone
Datadog
FastAPI
PyTorch
TensorFlow- LangSmith Tracing, evals and LLM observability in production
- ISO 42001 AI management systems
- EU AI Act Readiness and compliance by design
- Responsible AI Governance frameworks across use cases
- Auditability Traceable, reviewable AI decisions
- MLOps governanceCI/CD and standards across business units
- GCP Model Armor Prompt-injection and data-leak screening
- LangSmith LLM Gateway Rate and spend limits, PII redaction and audit logging
- Daytona Sandboxed code execution for agents
- Plugins Tools, connectors and skills bundled together, loaded by progressive disclosure
- Context engineering Progressive disclosure, memory and agent-readable knowledge bases
- Figma UX flows and high-fidelity prototypes
- Motion design Spring-based, interruptible interactions
- AI-assisted UI Prototyping with modern component and motion libraries
From idea to working systems
Selected work
An independent developer tool and personal open-source projects: a closer look at how I build agents, turn difficult inputs into useful data, and connect AI to real tools.
Credentials
Education
Otto von Guericke University Magdeburg · Machine Learning, Deep Learning & Reinforcement Learning
Bachelor of Computer Applications (BCA)
Guru Gobind Singh Indraprastha University, Delhi · Computer applications, software development & database systems
Certifications
Google Cloud · 2024
Professional Machine Learning Engineer
Building, deploying and fine-tuning machine learning models on Google Cloud.
Verify
HashiCorp · 2025
Terraform Associate (003)
Infrastructure as Code, automation and state management for AI and MLOps platforms.
Verify
Databricks · 2025
AWS Platform Architect
Configuring and optimizing Databricks on AWS for enterprise analytics and ML.
VerifyReferences
“Pawan Kumar demonstrated exceptional technical depth and initiative during his time at L3S. His contributions to machine learning-driven healthcare simulation and edge orchestration were delivered with a high degree of autonomy and rigor. His professionalism and dedication left a lasting impact on our research efforts.”
“Pawan is a rare talent who combines deep technical expertise with business acumen and cross-functional leadership. His GenAI solutions, cloud migration strategies, and CI/CD automation have tangibly improved our operational KPIs. He is a force multiplier for enterprise-scale AI initiatives and a cornerstone of our ML engineering team.”
Writing
On building and evaluating production GenAI systems, multi-agent architectures, RAG and agent tooling.
Let's talk.
Open to conversations about AI platforms, agentic systems and technical leadership roles.