Large Language Model Experts in Essen

• Training AI models, evaluating images and text UI/UX, turning the provided data into insights through OpenAI Feather as part of the machine learning workflow

Technologies: OpenAI Feather

Independent Consultant

• Advisory focus: E-Commerce, Digitalization, Automation, Pricing, Assortment & Merchandising, Platform Strategy
• Industries: Retail & Consumer Goods
• Subject Matter Expert for leading management consultancies
• Ongoing executi...

Tools: Fabric, AWS, dbt, Power BI, SQL, DWH, R, Python

• Supported customers in implementing an architecture design for extracting and preparing data
• Planned the design and implementation of the BI and DWH platform
• Ensured the scalability and performance of the data platform

• Acquires new B2B customers for an AI SaaS startup in the public tendering space and books product demos with IT decision-makers.
• Qualifies target customers based on a defined ideal customer profile, including discovery, needs analysis, and objection handling.
• Builds domain knowledge in public procurement (EVB-IT, German and EU procurement portals) for conversations on equal footing.

• Delivered MVP-focused full-stack architecture for a health-sector client: Vite/React frontend, backend services on Google Cloud Run, and Supabase for database plus IAM/authentication.
• Supported product requirements engineering and prioritized cost-aware workload placement, implementing browser-side/edge computation where feasible before moving logic to backend services.

• Title: Enhancing Syntactic Awareness in Transformer Language Models for Hindi Dependency Parsing
• Investigated syntactic knowledge captured by transformer language models (RoBERTa, XLM-RoBERTa) for Hindi dependency parsing, a morphologically rich and low-resource language.
• Developed structure-aware model variants (Struct_Roberta_hi, Struct_XLMR) by integrating a CNN-based parser network between transformer layers, inspired by the StructFormer architecture.
• Conducted extensive error analysis including label-wise, distance-based, direction-based, sentence length-based, and LVC/Non-LVC evaluations.
• Evaluated models on downstream NLP tasks (NER, POS tagging) using the IndicXTREME benchmark.

• Developed and optimised structured support workflows and evaluation procedures, applying consistent quality standards across high-volume operational tasks.
• Monitored performance metrics to identify systemic issues and drive targeted improvements — a skill directly transferable to LLM performance metric analysis.
• Managed escalations and maintained high accuracy and satisfaction standards in a fully asynchronous, remote-first environment.

Conventional Paradigm Test (CPT) – AI Evaluation & LLM Research

Development of an experimental evaluation approach to examine “paradigmatic closure” in Large Language Models — that is, the question of how far LLMs can recognize the basic assumptions, values, and limits of the paradigms within which they generate answers.

Design and testing of an additional approach to classic AI benchmarks that does not primarily measure factual correctness or task performance, but instead examines a model’s ability to recognize alternative perspectives, make implicit assumptions visible, and reflect on the limits of its own answer or interpretation framework.

Focus areas: development of evaluation criteria and test questions · LLM evaluation and comparative model analysis · prompt and response analysis · qualitative classification of model answers · study of epistemic compression and value leakage · benchmark and literature research · development of structured assessment and analysis methods

As part of CPT, existing AI evaluation approaches and benchmarks were analyzed, and a minimalist test protocol was developed that classifies model answers by response patterns such as DIRECT, CLARIFY, PLURALIST, REFUSE, and META-AWARE. TruthfulQA was used as the basis for experimental application and comparison with existing reference answers.

Technologies & Methods: Large Language Models (LLMs) · Generative AI · Prompt Engineering · AI Evaluation · TruthfulQA · Benchmark Analysis · Human-in-the-Loop Evaluation · Qualitative Content Analysis · Research & Literature Review

• Patent application: European patent drafting and prosecution
• LLM practicing: Developed LLM-based tools for automated patent data retrieval, applying Python scripting to accelerate technical reviews.

Conception, planning, and production deployment of two AI platforms for industrial research and engineering workflows, from use-case identification and requirements analysis through architecture decisions and build-vs-buy trade-offs to go-live.

• awRAG.io: Identification of the use case (fragmented knowledge base across distributed AI tools), definition of data requirements, architecture decision for a multi-tenant RAG-as-a-service platform with GDPR-compliant EU infrastructure and production-grade retrieval pipeline

• LAIERS.ai: Use-case definition (context loss in linear AI workflows), strategic product decisions on UX, cost structure, and multi-LLM orchestration, rollout of a spatial AI conversation platform with proprietary context management system LAICS

• LLMOps ownership: Quality assurance, pipeline optimization, security architecture (OAuth 2.0, SOC 2), and performance monitoring of both platforms in live production

• Core topics: LLM, RAG, vector databases, LLMOps, AI architecture strategy, cloud infrastructure, data sovereignty

• Strategic transition into AI Engineering through intensive mentoring and project execution.

• Developed MASX AI, an agentic AI platform integrating LangGraph, AutoGen, and RAG for geopolitical forecasting and real-time ETL.

• Designed and delivered functional AI prototypes for prospective clients showcasing applied expertise in multi-agent systems, real-time data pipelines, and LLM integrations.

• Re-architected a web application that had grown over years, modularizing tightly coupled components

• Reduced response times from several seconds to <50 ms

• Removed performance bottlenecks

• Simplified architecture so internal teams can implement features on their own again

• Conduct structured evaluation of LLM outputs using Content Review Standards (CRS) and AI safety frameworks.
• Assess responses across high-risk domains including violence and criminal facilitation.
• Assess responses across high-risk domains including hate speech and harassment.
• Assess responses across high-risk domains including suicide and self-harm.
• Assess responses across high-risk domains including regulated advice (medical, legal, financial).
• Assess responses across high-risk domains including misinformation and fabricated claims.
• Assess responses across high-risk domains including defamation and intellectual property.
• Assess responses across high-risk domains including child safety and sexual exploitation.
• Assess responses across high-risk domains including political and sensitive content.
• Apply youth-protection and age-appropriateness guidelines to prevent unsafe facilitation or restricted substance guidance.
• Classify prompts as adversarial, borderline, or benign based on contextual intent and risk analysis.
• Evaluate model behavior types including correct refusal, partial refusal, over-refusal, under-refusal, improper compliance, and ignorance-based outputs.
• Identify policy misapplications and user-intent misinterpretation patterns.
• Designed structured adversarial and borderline multi-turn conversation flows to stress-test AI boundary enforcement and reasoning stability.
• Identified failure modes including hallucination, unsafe compliance, excessive refusal, contextual drift, and inconsistent safety logic.
• Applied a structured four-dimension evaluation rubric covering accuracy & safety, relevance & completeness, clarity & structure, and tone & appropriateness.
• Provided structured feedback supporting supervised fine-tuning and reinforcement learning from human feedback processes.
• Rewrote unsafe or misaligned outputs into compliant, accurate, and helpful responses.
• Performed Persian ↔ English translation and translation validation of AI-generated content.
• Assessed semantic accuracy, contextual consistency, and safety alignment across languages.
• Identified mistranslations, cultural nuance issues, and cross-lingual policy inconsistencies.
• Recognized with the Above & Beyond Award – Q3 2025 for exceeding quality standards and embracing innovation.

• Lead Statistical Programmer for clinical trials
• Management of internal deliverables, programming resources, and external service providers (CROs)
• Development of automation systems and AI integration into statistical programming tasks
• Process analysis and KPI management for the leadership team