Overview Ueberblick Interested in Research and Engineering Interessiert an Forschung und Engineering

Ashish Rana

Building interpretable, reliable AI systems with research depth and production discipline. Ich entwickle interpretierbare, zuverlaessige KI-Systeme mit Forschungstiefe und Produktionsdisziplin.

I work across agentic LLMs, prototype-based explainable AI, multimodal learning, and scalable machine learning systems. My current research direction connects long-term memory in agentic LLMs, causal prototype-mechanism explanations, and production-minded LLM system design. Ich arbeite an agentischen LLMs, prototypenbasierter erklaerbarer KI, multimodalem Lernen und skalierbaren Machine-Learning-Systemen. Meine aktuelle Forschungsrichtung verbindet Langzeitgedaechtnis in agentischen LLMs, kausale Prototyp-Mechanismus-Erklaerungen und produktionsnahes LLM-Systemdesign.

Ashish Rana

About Profil

1 Reliable Agentic LLMs Zuverlaessige agentische LLMs 2 AI Interpretability Interpretierbarkeit von KI 3 Multimodal Representation Learning Multimodales Repraesentationslernen

Research profile Forschungsprofil

My research taste is centered on XAI, LLM systems, prototype-based explanations, and agentic memory. On the engineering side, I care about reproducible ML pipelines, complete deployed software with backend and CI/CD discipline, and AI software packages that are practical enough to run more than once.

Mein Forschungsinteresse liegt bei XAI, LLM-Systemen, prototypenbasierten Erklaerungen und agentischem Gedaechtnis. Auf der Engineering-Seite interessieren mich reproduzierbare ML-Pipelines, vollstaendig deployte Software mit Backend- und CI/CD-Disziplin sowie KI-Softwarepakete, die praktisch und verlaesslich nutzbar sind.

Languages Sprachen

English C1 Professional Working Proficiency
German B2 TELC B2 Certification
C1 in Progress
Englisch C1 Professionelle Arbeitsfaehigkeit
Deutsch B2 TELC-B2-Zertifizierung
C1 in Vorbereitung

Experience Erfahrung

Chronological View of Research Engineering, Data Science, and Software Systems Work.

Chronologische Ansicht von Research Engineering, Data Science und Software-Systems-Arbeit.

Professional Timeline

4 Years Industry Experience
2 Years Industrial Research
1 Year Explainable AI (XAI)
1 Year Reliable LLMs
Jun 2025 - Mar 2026 Full-Time
Research Engineer

NEC Laboratories Europe (NLE) · Heidelberg, Germany

  • Researched long-term memory in agentic LLMs and causal prototype-mechanism explanations.
  • Developed scalable real-time LLM systems for emergency call handling with FastAPI and CI/CD.
Long-term memory Causal prototypes System design Software releases Thesis supervision
Jan 2024 - May 2025 Full-Time
Research Associate

NEC Laboratories Europe (NLE) · Heidelberg, Germany

  • Built state-of-the-art interpretable deep variants of the Classification-by-Components approach.
  • Developed an interpretable health assistant web application with GLVQ and GTLVQ approaches.
Prototype-based learning Prototype building Software development
Apr 2023 - Dec 2023 Part-Time
Working Student trinamiX GmbH · Ludwigshafen, Germany
  • Developed data processing and code optimizations for a CNN-based deep learning pipeline.
  • Benchmarked transfer learning performance of EfficientNet with dynamic input-size biometric data.
PyTorch Hydra MLflow
Apr 2023 - Sep 2023 Part-Time
Working Student Institute for Enterprise Systems · Mannheim, Germany
  • Built data preprocessing and pipeline code optimizations for the BERT system pipeline.
  • Developed architecture improvements for smart meter time-series data.
Time-series data Transformers Data processing
Sep 2022 - Feb 2023 Part-Time
Working Student FARO Technologies · Korntal-Muenchingen, Germany
  • Conducted large-scale high-fidelity point cloud segmentation with RandLA-Net.
  • Finetuned RandLA-Net on synthetically generated data and benchmarked the improvements.
PyTorch Open3D-ML Synthetic data generation
Jun 2022 - Aug 2022 Full-Time
Data Science Intern ZEW · Mannheim, Germany
  • Developed web scrapers for multilingual data and translator wrappers for syncing datasets.
  • Conducted football competency statistical analysis with Difference-in-Differences models.
Selenium Beautiful Soup HuggingFace PyTorch
Aug 2019 - Sep 2021 Full-Time
Data Engineering Specialist

GE Healthcare · Bengaluru, India

  • Developed a predictive outlier detection proof of concept for MRI modality sensor data.
  • Built machine utilization metrics to measure extra healthcare system load for COVID-19 patients.
  • Developed and deployed terabyte-scale PL/pgSQL ingestion procedures on AWS Redshift.
Python R PL/pgSQL AWS Redshift
Jan 2019 - Jun 2019 Full-Time
Data Engineering Intern American Express · Bengaluru, India
  • Developed Prometheus and Grafana based monitoring and alerting tools for Couchbase DB VMs.
  • Designed, developed, and deployed parallel VM patching scripts to reduce downtime by four times.
Couchbase Python Grafana Prometheus
Jun 2018 - Jul 2018 Full-Time
Software Engineering Intern GE Digital · Bengaluru, India
  • Added a SAML re-authentication module and migrated the Enovia PLM codebase to REST architecture.
  • Conducted a case study of open-source static analysis tools on the Enovia Java codebase.
REST Java SAML

Beruflicher Werdegang

4 Jahre Industrieerfahrung
2 Jahre Industrieforschung
1 Jahr Erklaerbare KI (XAI)
1 Jahr Zuverlaessige LLMs
Jun 2025 - Maerz 2026 Vollzeit
Research Engineer

NEC Laboratories Europe (NLE) · Heidelberg, Deutschland

  • Forschung zu Langzeitgedaechtnis in agentischen LLMs und kausalen Erklaerungen durch Prototypen und Mechanismen.
  • Entwicklung skalierbarer Echtzeit-LLM-Systeme fuer die Bearbeitung von Notrufen mit FastAPI und CI/CD.
Langzeitgedaechtnis Kausale Prototypen Systemdesign Software-Releases Abschlussarbeitsbetreuung
Jan 2024 - Mai 2025 Vollzeit
Research Associate

NEC Laboratories Europe (NLE) · Heidelberg, Deutschland

  • Entwicklung interpretierbarer Deep-Learning-Varianten des Classification-by-Components-Ansatzes auf dem Stand der Technik.
  • Entwicklung einer interpretierbaren Webanwendung fuer Gesundheitsassistenz mit GLVQ- und GTLVQ-Ansaetzen.
Prototypenbasiertes Lernen Prototypenentwicklung Softwareentwicklung
Apr 2023 - Dez 2023 Teilzeit
Werkstudent trinamiX GmbH · Ludwigshafen, Deutschland
  • Entwicklung von Datenverarbeitung und Codeoptimierungen fuer eine CNN-basierte Deep-Learning-Pipeline.
  • Benchmarking der Transfer-Learning-Performance von EfficientNet mit biometrischen Daten dynamischer Eingabegroesse.
PyTorch Hydra MLflow
Apr 2023 - Sep 2023 Teilzeit
Werkstudent Institute for Enterprise Systems · Mannheim, Deutschland
  • Entwicklung von Datenvorverarbeitung und Codeoptimierungen fuer die BERT-Systempipeline.
  • Entwicklung von Architekturverbesserungen fuer Smart-Meter-Zeitreihendaten.
Zeitreihendaten Transformer Datenverarbeitung
Sep 2022 - Feb 2023 Teilzeit
Werkstudent FARO Technologies · Korntal-Muenchingen, Deutschland
  • Durchfuehrung grossskaliger, hochaufloesender Punktwolkensegmentierung mit RandLA-Net.
  • Feinabstimmung von RandLA-Net auf synthetisch erzeugten Daten und Benchmarking der Verbesserungen.
PyTorch Open3D-ML Synthetische Daten
Jun 2022 - Aug 2022 Vollzeit
Data-Science-Praktikant ZEW · Mannheim, Deutschland
  • Entwicklung von Web-Scrapern fuer mehrsprachige Daten und Uebersetzungs-Wrappern zur Synchronisierung von Datensaetzen.
  • Statistische Analyse von Fussballkompetenzen mit Difference-in-Differences-Modellen.
Selenium Beautiful Soup HuggingFace PyTorch
Aug 2019 - Sep 2021 Vollzeit
Data Engineering Specialist

GE Healthcare · Bengaluru, Indien

  • Entwicklung eines POC zur praediktiven Ausreissererkennung fuer MRT-Modalitaeten aus mehreren Sensordatenquellen.
  • Entwicklung von Kennzahlen zur Maschinenauslastung fuer COVID-19-Belastungsanalysen.
  • Entwicklung und Bereitstellung von PL/pgSQL-Ingestionsprozeduren im Terabyte-Massstab auf AWS Redshift.
Python R PL/pgSQL AWS Redshift
Jan 2019 - Jun 2019 Vollzeit
Data-Engineering-Praktikant American Express · Bengaluru, Indien
  • Entwicklung Prometheus- und Grafana-basierter Monitoring- und Alerting-Werkzeuge fuer Couchbase-DB-VMs.
  • Konzeption, Entwicklung und Bereitstellung paralleler VM-Patching-Skripte zur vierfachen Reduzierung der Ausfallzeit.
Couchbase Python Grafana Prometheus
Jun 2018 - Jul 2018 Vollzeit
Software-Engineering-Praktikant GE Digital · Bengaluru, Indien
  • Ergaenzung eines SAML-Re-Authentifizierungsmoduls und Migration der Enovia-PLM-Codebasis auf eine REST-Architektur.
  • Durchfuehrung einer Fallstudie zu Open-Source-Werkzeugen fuer statische Analyse auf der Enovia-Java-Codebasis.
REST Java SAML

Publications Publikationen

Recent industrial research contributions, mostly first-author, spanning agentic memory, interpretable AI, reinforcement learning, information retrieval, and scientific AI.

Aktuelle Beitraege aus der Industrieforschung, ueberwiegend als Erstautor, zu agentischem Gedaechtnis, interpretierbarer KI, Reinforcement Learning, Information Retrieval und wissenschaftlicher KI.

Earlier publications Fruehere Publikationen

Education Ausbildung

2021 - 2023

M.Sc. Data Science

University of Mannheim · Germany

GPA 2.4 (Good) German scale; approximately USA 3.1
2015 - 2019

B.E. Computer Engineering

Thapar Institute of Engineering and Technology · India

CGPA 9.51/10.00 Approximately DE 1.3 and USA 3.8
2013 - 2014

12th Grade

Army Public School, Meerut Cantt. · Uttar Pradesh, India

91%
2011 - 2012

10th Grade

Army Public School, Jalandhar Cantt. · Punjab, India

GPA 9.8/10.0
2021 - 2023

M.Sc. Data Science

University of Mannheim · Deutschland

Note 2,4 (Gut) Deutsche Notenskala; ungefaehr USA 3,1
2015 - 2019

B.E. Computer Engineering

Thapar Institute of Engineering and Technology · Indien

CGPA 9,51/10,00 Ungefaehr DE 1,3 und USA 3,8
2013 - 2014

12. Klasse

Army Public School, Meerut Cantt. · Uttar Pradesh, Indien

91%
2011 - 2012

10. Klasse

Army Public School, Jalandhar Cantt. · Punjab, Indien

GPA 9,8/10,0

Interests Interessen

1
Reliable Agentic LLMs Zuverlaessige agentische LLMs Memory, safety, and system behavior for long-running agentic AI. Gedaechtnis, Sicherheit und Systemverhalten fuer langfristig laufende agentische KI.

I work on long-term memory, decay-driven activation, and scalable LLM systems that remain useful under real operational constraints.

Ich arbeite an Langzeitgedaechtnis, zerfallsgetriebener Aktivierung und skalierbaren LLM-Systemen fuer reale Einsatzbedingungen.

2
AI Interpretability Interpretierbarkeit von KI Prototype-based explanations, causal mechanisms, and transparent model behavior. Prototypenbasierte Erklaerungen, kausale Mechanismen und transparentes Modellverhalten.

I build interpretable systems around Classification-by-Components, GLVQ/GTLVQ, and prototype-specific explanations for high-stakes domains.

Ich entwickle interpretierbare Systeme rund um Classification-by-Components, GLVQ/GTLVQ und prototypenspezifische Erklaerungen.

3
Multimodal Representation Learning Multimodales Repraesentationslernen Learning useful representations across language, vision, time-series, and structured signals. Nuetzliche Repraesentationen ueber Sprache, Vision, Zeitreihen und strukturierte Signale hinweg.

My experience spans CT image analysis, time-series transformers, point-cloud segmentation, information retrieval, and scientific claim verification.

Meine Erfahrung umfasst CT-Bildanalyse, Zeitreihen-Transformer, Punktwolkensegmentierung, Information Retrieval und wissenschaftliche Claim-Verifikation.

Contact Kontakt

Reach Me Via Email, LinkedIn, or My Website, and See My Work on Google Scholar.

Erreichen Sie mich per E-Mail, LinkedIn oder ueber meine Website; meine Arbeiten finden Sie auf Google Scholar.


Click to copy the address.

Klicken, um die Adresse zu kopieren.

Let's work together Lassen Sie uns zusammenarbeiten