Digital Factory AI Lead

Reshaping industry workflows by developing AI solutions and effective agentic systems.

Lead AI Engineer

moatless

Helping industry experts and researchers to navigate the sheer endless sea of Large Language Model (LLM) research & innovation because everybody deserves to find the best information.

Founder

Stealth

I was part of a 6 month government-funded Entrepreneurship program that enabled me to work on my own projects and ideas within the LLM space.

During this time, I gained experience in building LLM-powered backends with popular in-demand tools. The prototype I ended up building included the following:

- Fetching data from an external API & processing the response (Python)
- NoSQL database (MongoDB, Postgres)
- PDF Processing (PyPDF2, PyMuPDF)
- Document Extraction (Textract/Document Transformer)
- LLM Chain with multiple steps and prompt templates (Python, Guidance/LangChain)
- API endpoint for the processed data & the LLM outputs (FastAPI)
- Cloud Infrastruture (AWS)

Since I operated on a minimal budget, efficiency and cost reduction were a priority for me. Specifically, I focused on optimizing tasks like PDF processing, document understanding, information extraction and summarization.

Senior Machine Learning Engineer

A1 Telekom Austria AG

During my time as a Senior at A1, I was the DS lead for the company's most advanced NLP projects.

My work included

- Reporting results to several levels of management 
- Representing cutting edge Deep Learning & NLP at the company
- Pitching these topics to the C-Level and other influential exexutives
- Giving expert talks & presentations
- Participating in workshops
- Taking part in the recruiting process for full-time Data Science internships
- Guiding selected interns and assigning them roles in my projects

For a technical overview of my work, see below.

Senior Data Scientist

I was in charge of a Social Media Listening project for Facebook, Twitter and Online Newspapers which included the following

- SQL & MongoDB databases
- Web scraper for the newspaper(s)
- Processing Facebook & Twitter API data
- Transfer Learning & Language Model Finetuning on in-domain texts
- Multiclass and Multilabel Text Classifiers based on Embeddings and Transformers (the original PyTorch implementation of BERT)
- Tableau Dashboard

On top of that, I was developing intents & entities for customer text channels, including e-mail, web search, specialized chatbots (e.g., billing), the aforementioned Social Media data and several smaller Text & Sequence Classification projects.

I then moved on to explore speech channels and became the person in charge of Speech-to-Text at the company, which included 

- Processing short IVR audio recordings
- Processing long(er) call center recordings 
- Labeling pipeline with custom Prodigy scripts
- Developing a subtitle matching algorithm for aligning audio with text & filtering out errorenous subtitles
- Developing Hybrid HMM-TDNN Speech Recognition models with Kaldi
- Switching to (back then freshly released) bleeding edge libraries & models for End-to-end Deep Learning, such as NVIDIA NeMo & Facebook FAIR
- Improving performance on Austrian dialects & in-domain recordings
- Adapting the in-house Classification models mentioned earlier for the text output of the speech channels

In addition, I was able to work on PoCs in the areas of Document Extraction, Document Classification, Entity Extraction & Question Answering.

Data Scientist

Universität Wien

Research in Cognitive & Economic Psychology:

- Dual process theory (Type I/II thinking)
- Metacognition theory
- Experimental assessment of cognitive biases
- Leveraging aggregated mouse cursor movement data

Junior Researcher

Coursera

Data Science Education

As the founder of LLM Watch, my goal is to curate the most relevant Large Language Model (LLM) research and make it accessible to a broad audience. 

I distill complex technical topics into summaries that respect your time. 

Before embarking on this journey, I led Deep Learning R&D at one of Europe's largest telecom companies. There, I spearheaded numerous AI projects with a focus on Natural Language Processing (NLP) and Automatic Speech Recognition (ASR). 

With more than 6 years of experience in AI, I was there before the hype. 

I try to offer a healthy balance between shiny new things and what really matters.

Pascal Biese

Yann LeCun's first big paper after he left Meta. Here's what they did:

For years, training Joint-Embedding Predictive Architectures (JEPAs) has been more art than science. Stop-gradients, teacher-student networks, carefully tuned EMA schedules – a delicate dance of heuristics to prevent representation collapse. But what if there was a principled way?

Self-supervised learning promised to unlock the power of unlabeled data, yet state-of-the-art methods like DINOv2 and I-JEPA rely on complex recipes with dozens of hyperparameters. Each new dataset or architecture requires extensive tuning. Even worse, the training loss often doesn't correlate with downstream performance, making model selection nearly impossible without expensive labeled evaluation.

The core problem: we've lacked a theoretical understanding of what makes good representations. LeJEPA changes this by proving a fundamental result – isotropic Gaussian embeddings uniquely minimize worst-case risk across downstream tasks. 

Tho achieve this, they introduce SIGReg (Sketched Isotropic Gaussian Regularization), a novel objective that enforces this optimal distribution using random projections and characteristic function matching. The beauty lies in its simplicity: linear complexity, bounded gradients, and provably correct by construction.

LeJEPA eliminates stop-gradients and teacher-student networks entirely while achieving competitive performance – 79% on ImageNet-1k with a frozen ViT-H/14. And here's the best part: it works out-of-the-box across 60+ architectures and diverse domains - in-domain pretraining with LeJEPA on small datasets (even just 1,000 samples) outperforms transfer learning from massive frontier models trained on billions of images.

This could fundamentally shift how we think about foundation models. Instead of relying on ever-larger generic models, practitioners in specialized domains can now train competitive models directly on their data with minimal hyperparameter tuning. 

The fact that LeJEPA's training loss actually correlates 99% with downstream performance also solves the critical model selection problem that has plagued SSL research.

What do you think? Is Yann onto something?
↓ 
𝐖𝐚𝐧𝐭 𝐭𝐨 𝐤𝐞𝐞𝐩 𝐮𝐩? Join my newsletter with 50k+ readers and be the first to learn about the latest AI research: llmwatch.com 💡

AI Lead at PwC <​/​> Daily AI highlights for 80k+ experts 📲🤗

LeJEPA: Provable and Scalable Self-Supervised Learning

Microsoft just announced the "Era of Agentic Organization". Here's what that means:

As LLMs scale their reasoning capabilities, we need new techniques: sequential thinking is too slow, and parallel thinking wastes compute by running independent traces that can't learn from each other mid-process. The industry needs a paradigm that combines the adaptivity of sequential reasoning with the speed of parallelization.

For this reason, Microsoft Research developed AsyncThink. AsyncThink is a new approach where a single LLM learns to play two roles simultaneously: an organizer that dynamically structures the reasoning process through Fork and Join operations, and workers that execute sub-queries concurrently. 

Their key contribution is that the model learns via reinforcement learning to optimize its own thinking structure - deciding when to split problems, which sub-queries to pursue, and when to merge intermediate results.

The results? 28% lower inference latency compared to parallel thinking while maintaining higher accuracy on mathematical reasoning. Models trained only on simple countdown problems successfully generalize their asynchronous thinking capabilities to completely unseen domains like Sudoku and complex math problems.

The more tasks AI is supposed to solve within an organization, the more important it becomes to structure all of this in a scalable and controllable way. AI agents need to learn to organize their reasoning like skilled project managers, adaptively coordinating multiple thinking processes based on the problem at hand. 
↓ 
𝐖𝐚𝐧𝐭 𝐭𝐨 𝐤𝐞𝐞𝐩 𝐮𝐩? Join my newsletter with 50k+ readers and be the first to learn about the latest AI research: llmwatch.com 💡

The Era of Agentic Organization

Can we train billion-parameter models without gradients? Oxford researchers just showed it's possible - at 100x the speed.

Evolution Strategies have always been an elegant alternative to backpropagation. They don't need differentiability, handle noisy objectives gracefully, and parallelize beautifully. But there's been a fundamental problem: they simply don't scale to modern model sizes.

University of Oxford's new EGGROLL algorithm mirrors what made LoRA successful for fine-tuning - instead of perturbing all parameters with full-rank matrices, use low-rank approximations. For each population member, sample matrices A and B to form a rank-r perturbation AB^T rather than a full matrix E. This cuts memory from mn to r(m+n) per layer and reduces forward pass compute proportionally.

What makes this work is subtle but powerful: while each perturbation is low-rank, the final parameter update averages across the entire population, yielding a full-rank update when the population is large enough. The researchers prove this approximation converges to full-rank ES at an O(1/r) rate - faster than typical central limit theorem rates due to the symmetry of the problem.

The results validate the approach across three domains. In pure integer pretraining of recurrent language models, EGGROLL enables stable training with population sizes exceeding 250,000 - orders of magnitude larger than prior work. For LLM reasoning tasks, it matches or exceeds GRPO while maximizing hardware utilization. And critically, it achieves near-inference throughput speeds, making gradient-free optimization genuinely practical at scale.

This opens real possibilities for training systems that can't use backpropagation: end-to-end neurosymbolic architectures, models with discrete components, or deployment scenarios where integer-only computation is essential for efficiency.
↓ 
𝐖𝐚𝐧𝐭 𝐭𝐨 𝐤𝐞𝐞𝐩 𝐮𝐩? Join my newsletter with 50k+ readers and be the first to learn about the latest AI research: llmwatch.com 💡

Evolution Strategies at the Hyperscale

We might be systematically underestimating LLMs.

We've been told that reinforcement learning is essential for frontier reasoning capabilities. Massive compute budgets, curated datasets, reward models - the whole apparatus. 

But new research from Harvard University challenges this expensive paradigm with a surprisingly simple question: can we just sample better from base models?

The paper introduces a training-free algorithm that achieves reasoning performance matching or exceeding GRPO (the standard RL method) across mathematics, coding, and science benchmarks. 

The key insight revolves around the "power distribution" - sampling from p^α instead of p. Unlike low-temperature sampling which greedily averages future token likelihoods, power sampling accounts for entire future completion paths. 

This seemingly subtle difference makes a big difference: it naturally upweights tokens with fewer but higher-likelihood futures, essentially planning ahead rather than getting trapped in dead-ends.

Using Metropolis-Hastings MCMC to approximate this distribution, the researchers show that Qwen2.5-Math-7B can reach 74.8% on MATH500 without any additional training - nearly matching the 78.5% from full RL posttraining. 

On out-of-domain tasks like HumanEval, it outperforms by up to 20 percentage points. Perhaps most remarkably, unlike RL which suffers from diversity collapse, this method maintains strong pass@k performance across multiple samples.

The takeaway is that base models may already contain the reasoning capabilities we're spending enormous resources to extract through RL. This would open up a new axis for inference-time compute scaling that's accessible to anyone with a base model - no training infrastructure, curated datasets, or verifiers required.

We're still learning how to properly query the "intelligence" that already exists in our models.
↓ 
𝐖𝐚𝐧𝐭 𝐭𝐨 𝐤𝐞𝐞𝐩 𝐮𝐩? Join my newsletter with 50k+ readers and be the first to learn about the latest AI research: llmwatch.com 💡

Your Base Model is Smarter Than You Think

ThinkIT Data Solutions

Designing and operating datacenter infrastructure for linear broadcast video and associated telemetry services.

Principal IP Video Engineering Consultant - Cox Communications Video Engineering & Broadcast Ops

Mission Critical :: Cloud Native Solutions - Networking - Storage - Complex Telemetry Solutions - Managed Services

Founder/CxO

Implemented and supported Next Generation Data Center virtualization infrastructure based on Oracle/SUN technologies.  Computing stack is composed of Oracle 6000 Blade Chassis housing T4-1B SPARC Blade Servers running Oracle VM Server on Solaris 10/11.  Hundreds of Solaris LDOMs and zones provide highly available applications clustered via VCS (Veritas Cluster Service) and consist of custom “in-house” developed (predominately C++ & Java) applications, MQ and Tuxedo message brokering services, Oracle DB’s, and WebLogic, Tomcat, and JBoss application servers.  Administrative tasks including converting legacy run control scripts to SMF, creating IPS packages for applications, and developing scripts for the automated provisioning of virtual resources.  The networking stack is composed of 1GbE, 10GbE, 40GbE NEMs modules and 8Gb Fibre Channel.  Ethernet connectivity is via Cisco Nexus 2K’s, 5K’s, and 7018 core switches and Fibre Channel traverses Brocade DCX’s and 5300’s.  The storage stack is composed of NetApp filers providing NFS and Block storage services.  The Oracle virtualized environment provides a 24x7 highly available infrastructure for SITA’s core mission critical services to the airline transport industry and is distributed across 3 data centers and 2 continents.

Solaris SME and Virtualization Architect Consultant - SITA (sita.aero)

Helping businesses realize the significant benefits of virtualization and private cloud technologies thru elegant Infrastructure as a Service Architecture.  We also perform Virtual Infrastructure Optimization, SAN & Network Assessments, and SAN extension.

CTO/Chief Systems Architect

Short term engagement providing SAN extension, Virtualization, and Infrastructure Architecture Consulting for Geo Spatial Systems Data Center migration projects.

Principal Systems Architect Consultant to Sev1Tech

Short term engagement providing virtualization infrastructure architecture consulting services for remote data center and distribution facilities consolidation.

Principal Systems Engineering Consultant to Harris Teeter Corp.

Provide solutions architecture for Enterprise Data Center Operations Enterprise Storage Group.  Provide storage network architecture, management and monitoring, operations strategic guidance, and oversight.  Provide strategy to leverage current and implement new technologies to transition CBP’s enterprise storage environment to a services / cloud environment.  Specific HW involved: HDS VSP, USPV, NetApp 6030/v6080, EMC VMax, DMX-4 storage subsystems;  Oracle Exadata and SUN 4540 series NAS storage platforms; Sepaton VTL and Quantum iScalar tape libraries; and Brocade DCX, 48K directors, McData i10K & 6140 directors in a FICON environment. Utilize 8GB Fibre Channel, FibreChannel routing and FCIP.  Hosts consist of IBM z10 series mainframes, RedHat, Solaris, AIX and Windows based hosts in a heavily virtualized environment.  Tools include Virtual Instruments, NetApp SANscreen, Implementation and configuration of EMC’s ECC (now known as Ionix) & SMARTS/SIA, NetApp Operations Manager & FilerView, HDS HiCommand & Storage Navigator, and many others.

Principal Systems Architect Consultant to U.S. Customs & Border Protection

Provide storage network engineering solutions with a specialization in Enterprise Meta SAN design, routing and distance extension over IP (FCIP) and multi-vendor fabric integrations.  Very experienced with Main Frame FICON environments and FICON/Open Systems intermixed fabrics.  Experienced with File Virtualization and Global Namespace Design.  Experienced with HDS, IBM, EMC storage arrays, SUN "thumpers", StorageTech, Quantum tape libraries and VTL’s and NetApp filers.  Very experienced in the management challenges in multi-vendor fabrics, most vendors management applications and API’s.  Very familiar with both versions of the new Ethernet standard under proposal (CEE/DCE) for support of FCoE (FibreChannel over Ethernet).

Principal Engineering Consultant to Brocade Communications

•	Develop and execute load, stress, and performance tests using Load Runner of Raytheon’s enterprise manufacturing execution system (MES), Visiprise Manufacturing.
•	Develop and execute load, stress, and performance tests using Load Runner and SAP GUI scripting of Raytheon’s SAP based Trade Compliance System via traditional SAP GUI client as well as SAP’s Netweaver Portal. 
•	Reverse engineered custom DLLs and rewrote algorithms and program logic using built-in Load Runner functions and ANSI C to simplify program maintenance and portability.
•	Performed targeted performance testing and analysis for WebLogic and Oracle tuning and optimization for MES and various SAP portal based applications.
•	Numerous Perl programming development projects

Sr. Software Engineer Consultant to Raytheon Missile Systems

•	Provide network engineering services for Mentora’s Hosting customers.
•	Provide configuration and support for routing, switching and F5 Load Balancing devices and Linux, Solaris and Microsoft servers.
•	Reengineered F5 load balancing architecture to accommodate customer’s (Inovis) request to more granularly manage and monitor customer network traffic.

Principal Network Engineer Consultant to Mentora Group

Wachovia

Perform Load, Stress and Performance Testing for Wachovia's Online Banking channel.

Sr. Performance Test Engineer

Tenacity Solutions Incorporated

•	Technical lead and supervisor for Internet Operations and Web hosting groups within TCS.
•	Manage multiple government websites backups using NetBackup and a multitude of different hardware platforms. 
•	Oversee adherence to 3rd party hosting provider SLA’s.  
•	Consult with new customers to facilitate requirements gathering.  
•	Provide technical input for new business RFPs.  
•	Develop custom automated solutions for customer work processes and management of websites and distributed applications.  
•	Design and implement custom solutions for complex routing, firewalls, VPNs, and load balancing.  
•	Provide support for Treasury Enterprise Directory Services architecture consisting of x.500 and LDAP.

Principal Network Engineer at U.S. Treasury (TCS)

SOZA

Sr. Network Engineer for Internet E-Commerce Team

Designed performance testing environments for e-commerce environment utilizing Mercury Interactive's LoadRunner load testing suite.

Built out east and west coast nodes of e-commerce sites for the first online AND geographically load balanced presence of the U.S. Mint's catalog for coin sales and children's sites hosted in Sterling, VA and Burbank, CA QWEST datacenters

Design and implemented LAN infrastructure for U.S. Mint's 801 9th st. newly constructed location utilizing Foundry Networks BigIron Core switches, Cisco 7214 routers and Foundry FastIron workgroup switches.  Axent Raptor application gateway firewalls.

Sr. network engineer supporting nationwide WAN connecting all U.S. mint facilities including Fort Knox.

Principal Systems Architect at U.S. Mint

US Navy

The primary mission of Mobile Inshore Undersea Warfare Unit 210 is to enhance overall Force readiness by providing a rapidly deployable force capable of providing surface and subsurface surveillance in AOAs, harbors and approaches, roadsteads, straits, anchorages, offshore economic asset areas, and other militarily significant inshore areas throughout the world.

Secondary mission capabilities include command, control and communications, control of crafts of opportunity, control of ship movement within harbors and/or anchorages, and as an operational test and evaluation unit. A MIUWU is currently the only U.S. Navy platform specifically designated to fulfill this mission.

Electronics Technician (ET2) - Mobile Inshore Undersea Warfare Unit 210 - Reserve Force

Intelsat

•	Primary network and telecommunications systems support for Satellite Operations Dept.  Supported a multi protocol LAN/WAN environment consisting of Novell 4.11, NT4.0, VMS/DECNet, Solaris 2.6 & 7, HP/UX 10.20, RTE-A on HP A900 & 1000 servers, TimePlex multiplexors, ACT SDM series FRADs (for voice over frame), and Ascend B-STDX 8000 Frame Relay switches.  
•	Provided maintenance and support for satellite earth station equipment (700W TWT Ku-band & 3KW Klystron MCL HPAs, U/D Converters, Satellite Modems, Antenna and control systems).  
•	Performed integration and testing for new systems deployment and architecture upgrades.  
•	Developed and maintained systems engineering documentation.

Sr. Communications System Engineer

Comsat World Services

There are 6 tracking, telemetry, command, and monitoring (TTC&M) stations around the world and by the dedicated staff that support their continuous operation.  INTELSAT's TTC&M stations are in operation at: Beijing, China; Clarksburg, Maryland (USA); Fucino, Italy, Paumalu, Hawaii (USA); Perth, Australia; and Raisting, Germany.  Signatories in these countries are under contract to provide facilities and services for INTELSAT's use.  Thousands of earth station anetnnas of all sizes, from very large 18-meter gateway antennas to very small (1.8-meters or less) VSAT antennas access the INTELSAT system. 

Clarksburg TTC&M Earth Station:  Primary operations and technical support for all satellite telemetry, commanding, communications monitoring, TDMA reference terminal, and IDR communications equipment.

Performed certification testing on all local earth station antennas, transmit and receive subsystems, and computer and control equipment.

Performed equipment installations, custom cable fabrication and micro/miniature multi layer circuit board repair and soldering.

Primary technical support for 3KW Klystron and 700 Watt TWT HPAs.

Primary support for PC based automated wave guide switching system consisting of GE FANUC PLCs connected via RS-485 busses to a PC based program track and interface console.

Assisted in the design & implementation of a remote equipment enclosure temperature monitoring system consisting of remotely located thermistors on custom designed/built circuit boards connected via an RS-485 bus to a central PC w/a graphical display and alarm alerting features.

Performed miniature component soldering and multi-layer circuit board repair.

Sr. Electronics Technician

•	USS John F. Kennedy (CV-67):  Supervisor and Electronics Quality Control Inspector for a 12 technician work center.  
•	Primary technical support for intelligence (TRE, TACINTEL, JDIS, FIST, WWMCS), tactical (TADIX/OTCIX) and messaging (NAVMACS/Fleet Broadcast) UHF and SHF (X-Band) satellite communications systems.  
•	Installed and certified Fiber Optic, CAT 5, and RF/IF cabling.  
•	Performed Novell 3.12 and Unix/SUNOS 4 systems administration, communications equipment maintenance management, and systems training.  
•	Instructed Electronics and Satellite Communications Systems theory, operations, and maintenance training classes to Electronics support staff.
•	Performed miniature component soldering and multi-layer circuit board repair.

Electronics Technician (ET2) - USS John F Kennedy (CV-67)

Seasoned, accomplished infrastructure architect with 20+ years experience in distributed systems and high-volume, multi-platform environments. Exemplary architect, who can manage change, mitigate risk, infuse new ideas and deliver solid business results, taking enterprise system performance and productivity to the next level. Creative problem solving and troubleshooting skills complemented by the ability to communicate the problems and solutions to the business clearly and concisely.

My career started in 1984 working on Electrical, Fire Alarm and Motor Control Systems.  In 1990 I began a progression to Advanced Electronics, Terrestrial & Satellite Communications, Software Engineering, Systems Performance Testing, Distributed Systems Infrastructure and Network Engineering.  Currently designing technology-led business transformation solutions to provide for fully autonomous closed-loop operations for enterprise clients.

Specialties: RF Terrestrial & Satellite Communications, Systems Integration/Migration, Systems Architecture, Network Infrastructure, Software Development, Data Center Operations, Technology Planning, Project Management, Technical Support, Customer Relationship & Vendor Management

taplio

The Battle

Pascal Biese

Greg Phillips

Followers

Posts

Engagements

Likes

Performance ScorePascal Biese's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Pascal Biese

Greg Phillips

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Pascal Biese's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScorePascal Biese's LinkedIn Performance