Skip to content

Open to leadership and Individual Contributor roles · Germany

Builder of AI Products, Teams, and Systems

I take ideas from whiteboard to working software, owning the product decisions, the engineering, and the team that ships it. Eight years across startups and enterprise, always focused on what gets used.

View ProjectsGet in Touch

About

From code to customers to teams

Years of learning what it takes to make software matter, told in four chapters.

Foundations

I started in software engineering and embedded systems: writing backend services in Java, wiring up databases, learning how production code behaves when real users depend on it. That grounding still shapes how I evaluate every technical decision.

Evolution

A master's in computational modeling pulled me into machine learning and computer vision. At Dennemeyer and Daimler Truck, I moved from notebooks to pipelines, building NLP classifiers analysts could trust and time-series models R&D could plan around.

Expansion

The more I shipped, the more I cared about the parts around the model: who uses it, why it matters, whether anyone adopts it. Customer conversations, product tradeoffs, and team dynamics became as important as the architecture.

Today

I build products, teams, and systems end-to-end, using agentic AI, LLMs, and modern AI tooling in how I prototype, ship, and lead. At AICU I took a B2B SaaS platform from first interviews to paying customers. I still write code, still sit with customers, still hire and coach engineers. The best outcomes need all of it.

How I Build

Principles earned from shipping

Not a manifesto. Just what I've learned holds up when the work gets hard.

Build for Adoption

A model no one uses is just an expensive experiment. I start with the user, work backward to the system, and measure success by whether people change how they work.

Stay Close to the Work

Good calls come from reading the logs, sitting in the customer call, and understanding why the deploy failed at 2am. Not from reviewing slides about the deploy.

Own Outcomes

Shipping is the midpoint. I stay accountable for whether the system performs, whether the team can maintain it, and whether the business actually benefits.

Simplicity Scales

If I can't explain how a system works to the person operating it, it's not ready. Clear architecture outlasts clever architecture.

Skills

Skills & Technologies

Aligned with how I work: core stack, AI systems, infrastructure, architecture, leadership, and tooling.

Core Stack

PythonDjangoFastAPINext.jsAngularPostgreSQLBash

AI & ML

Computer VisionOpenCVPyTorchLLMsAgentic AIRAGNLPTransformersMLflow

Infrastructure

AWSAzureKubernetesDockerTerraformCI/CDGitHub ActionsJenkinsPrometheusGrafanaPosthog

Architecture

Multi-tenant SaaSSystem DesignScalable APIsData PipelinesModel ServingInference Optimisation

Leadership

Founding Team Hiring & ScalingCross-Functional Team CultureAI/ML Roadmap DevelopmentPrototype-to-Production DeliveryTechnical KPI DefinitionStakeholder Communication

AI Tooling

CodexCursorGitHub CopilotClaude

Languages

English (C2)German (B1)Hindi (Native)

Projects

Problems owned, outcomes delivered

Each project answers what mattered, what I owned, how I approached it, and what changed.

AI Workflow Platform

Full Stack
Problem
Industrial teams were running AI experiments in isolation. Useful in demos, but disconnected from the workflows they actually needed to change.
What I Owned
Owned product direction, customer discovery, technical architecture, and engineering hiring. Led LLM prototyping, backend, frontend, and AI workstreams.
Approach
Ran customer interviews before writing production code. Prototyped with LLMs to test hypotheses cheaply, then committed to a multi-tenant SaaS architecture only after validation. Chose depth over breadth in early features.
Impact
B2B SaaS from 0 to paying customers. Engineering team scaled from 2 to 15 across backend, frontend, and AI.
PythonDjangoLLMsKubernetesAWSMLflow

Battery Health Prediction

AI & ML
Problem
Electric truck fleet operators couldn't reliably forecast battery degradation. Maintenance planning and R&D investment depended on guesses, not data.
What I Owned
Owned use-case definition with R&D stakeholders, model development, and MLOps integration into existing Azure pipelines.
Approach
Prioritized interpretable time-series models over black-box complexity so engineering teams could trust outputs. Integrated training and monitoring into Azure MLOps rather than running a parallel workflow.
Impact
~90% improvement in prediction accuracy. ~20% reduction in model training time through pipeline optimization.
PythonAzure MLOpsSparkScikit-learn

Patent NLP Analytics

AI & ML
Problem
Patent analysts spent hours on repetitive document classification: high-volume, rules-heavy work that slowed down client deliverables.
What I Owned
Owned end-to-end NLP pipeline: data preparation, model selection, evaluation framework, and integration into existing analytics workflows.
Approach
Started with classical baselines, moved to transformers only where the accuracy gain justified the operational cost. Built evaluation around analyst trust, not just leaderboard metrics.
Impact
~95% classification accuracy on the patent NLP pipeline. Existing ML workflows improved by ~10% through systematic experimentation.
PythonTransformersNLPScikit-learn

Enterprise Backend Platform

Backend
Problem
Enterprise clients needed faster information retrieval across large relational datasets, and leadership wanted a credible path toward graph-based storage.
What I Owned
Owned backend feature delivery, relational-to-graph migration feasibility study, and performance improvements across production services.
Approach
Shipped incremental Spring Boot improvements while running a structured feasibility analysis for Neo4j migration, choosing evolution over big-bang rewrite. Used Kafka to decouple retrieval bottlenecks.
Impact
8% improvement in client information retrieval time. Delivered a clear migration recommendation leadership could act on.
JavaSpring BootPostgreSQLNeo4jKafka

Experience

Increasing scope, same standard

Each role added more ownership, from backend features to products, teams, and systems.

  1. Founder & Product Lead (CTO)

    AICU GmbH

    • Built B2B AI SaaS from 0 to paying customers, owning product vision, customer validation, and technical roadmap.
    • Scaled engineering from 2 to 15; defined hiring standards, team ownership, and delivery practices.
    • Shipped LLM prototypes to test product hypotheses with customers before committing engineering resources.

  2. Senior Data Scientist

    Daimler Truck AG

    • Defined ML use cases for battery health prediction with R&D and engineering stakeholders.
    • Delivered time-series models with ~90% accuracy improvement; integrated pipelines into Azure MLOps.
    • Reduced model training time by ~20% through pipeline refactoring and workflow optimization.

  3. Data Scientist

    Dennemeyer Octimine

    • Built NLP classification system for patent analytics with ~95% accuracy.
    • Improved existing ML pipelines by ~10% through structured experimentation and optimization.
    • Owned master thesis research applied directly to production patent data workflows.

  4. Associate Software Engineer

    AFour Technologies

    • Built enterprise backend features with Java and Spring Boot for production client applications.
    • Led feasibility work for relational-to-graph database migration.
    • Delivered 8% improvement in information retrieval time through architecture changes.

Contact

Let's talk about what you're building

Open to roles where I can own products, teams, and technical execution, at startups or in enterprise.

Email

singh.shivangi5march@gmail.com

LinkedIn

linkedin.com/in/shivangi-data-scientist

GitHub

github.com/Shivangi0503

Medium

shivangi-singh.medium.com