React Native
Running Large Language Models Fully Offline on Mobile with React Native
Paolo Pecis explores fully offline speech-to-text and LLM summarization on iOS and Android with React Native, whisper.rn, llama.rn, and the SoloAI app.
From the people building it
The latest thinking.
- Agent identity
Everything you need to know about Amazon Bedrock AgentCore Identity
Authentication and authorization for AI agents on AWS. Who is allowed to call your agent, and how your agent proves itself to everything it calls.
Hardware benchmark
Explore the collection
38 articles
- React Native
Running Large Language Models Fully Offline on Mobile with React Native
Paolo Pecis explores fully offline speech-to-text and LLM summarization on iOS and Android with React Native, whisper.rn, llama.rn, and the SoloAI app.
Agent identityEverything you need to know about Amazon Bedrock AgentCore Identity
Authentication and authorization for AI agents on AWS. Who is allowed to call your agent, and how your agent proves itself to everything it calls.
- Hardware benchmark
Scoring 42 Million Protein Variants a Day on AWS Trainium2
A native PyTorch benchmark showing EvolutionaryScale ESMC-300M running on AWS Trainium2: 490.8 variants per second on a single trn2.3xlarge — approximately 42.4 million protein variants scored per day at approximately $1.26 per million, 60% below H100 in this fixed-shape D2Deep benchmark.
Product DesignA Prompt, Pointed the Other Way
Everything is buildable now. Every team I meet lives in the same climate: more ideas than quarters to test them in, more tools than problems, roadmaps that read like wish lists because nothing on them is technically…
GrokBenchmarking Grok 4.3 on Amazon Bedrock Mantle vs. the xAI API: The Engineering Walkthrough with Loka
The main post covers what we found: Grok 4.3 lands in the same 96–97% accuracy band on GSM8K whether you reach it through Amazon Bedrock Mantle or the first-party xAI API, and low reasoning effort captures most of the…
Provider benchmarkThe Many Paths to Grok 4.3
A public-facing benchmark story on Grok 4.3 through Amazon Bedrock Mantle and the xAI API, focused on accuracy, latency, cost, reliability, and enterprise deployment tradeoffs.
BioinformaticsContributing to Open-Source Bioinformatics: Our Experience at the nf-core Hackathon from Medellín
If you work anywhere near data-heavy science — genomics, metagenomics, proteomics — you’ve probably hit the same wall: building reliable, reproducible data pipelines is hard. Nextflow is an open-source workflow engine…
NextflowFrom Proteins to Pipelines: How Open-Source Nextflow Tools Are Accelerating Drug Discovery
Modern computational drug discovery depends on complex workflows that connect AI models, structural biology tools, databases, and quality-control steps into reproducible pipelines. While many organizations have access…
Computer VisionComputer Vision at Scale on SageMaker Jobs: Four Decisions
Training a vision model on millions of images is, more than anything, a data problem wearing an ML costume. The architecture can many times be the least interesting part. Everything around it is usually more fun: how we…
Open SourcePushing Open-Source TTS Models to Their Limits: Six Paradigms, 14 Models, One Production Reality Check
In the digital audio subset of AI, the questions currently on everyone’s mind are
Amazon Bedrock GuardrailsAWS Bedrock Guardrails — Implementing an Allowlist
When building AI agents with AWS Bedrock, guardrails are your first line of defense for keeping conversations on topic. A common requirement is to restrict an agent to a pre-defined strict allowlist, blocking everything…
Speech synthesisPushing Open-Source TTS Models to Their Limits
Fourteen open-source text-to-speech models across six emotion-control paradigms, benchmarked against AWS Nova Sonic v2 on naturalness, expressiveness, latency, and licensing.
AlphaFoldRunning Six Open-Source Cofolding Models on AWS: Lessons Learned from a Compute Perspective
What to know before you scale cofolding jobs from five structures to ten thousand.
PythonWatching a Python-to-Rust rewrite was painful enough to build this
How we built an agent that ports Python ML code to edge hardware for space.
AIBenchmarking GPT-5.5 on Amazon Bedrock vs. the OpenAI API: The Engineering Walkthrough with Loka
The main post covers what we found: GPT-5.5 on Amazon Bedrock matches OpenAI API accuracy and comes out faster on every latency and throughput dimension we measured. This post covers how we ran it. If you want to…
Model evaluationHow Loka Evaluates and Builds with Frontier Models on AWS
A comparison of GPT-5.5 through Amazon Bedrock and the OpenAI API across answer quality, latency, throughput, and reliability, followed by two production-shaped copilot builds.
Accelerator benchmarkRunning Hugging Face Carbon on AWS Trainium2
A practical benchmark showing Hugging Face Bio Carbon running on AWS Trainium2 with NxD Inference, covering what worked, what we measured, and why it matters for bio and HCLS teams moving open models into production.
AWS TrainiumRunning Hugging Face’s Carbon on AWS Trainium2 with NxD Inference
Hugging Face Bio’s Carbon release is interesting for two reasons at once. First, it is a biology model. Second, it is not an infrastructure outlier. Many genomic models come with custom architectures, specialized…
Data LineageData Governance in Practice
Lineage for a CDC Pipeline on AWS
AWSBeyond Static Rules: Building Agentic Content Evaluation Systems on Amazon Bedrock
How to turn scattered guidelines into adaptive workflows that score, explain, and improve content at scale
ElevenLabsIA.COMédia: A Loka-made Lab for Theatre with Live AI Actors
A Portuguese playwright wanted to push AI beyond the assistant role and onto the theatre stage. Here’s how LOKA built a live system that brings a cast of AI actors to perform alongside humans.
AWSThe Agent Assembly Line
Productionizing Agentic Use Cases in Weeks, Not Months
ArceeDeploying Trinity-Mini-DrugProt-Think on Amazon SageMaker AI
If you work in regulated domains (Healthcare, Life Sciences, Finance) you routinely hit constraints that break the default “just call a hosted API” approach:
Reinforcement learningPost-Training an Open MoE Model to Extract Drug-Protein Relations
An ablation-driven study of GRPO-style reinforcement learning with LoRA on Trinity Mini for biomedical relation extraction, covering the training choices that materially changed performance.
AndroidClean Architecture with Strategy Pattern in Kotlin
When building stable and maintainable software, one of the biggest challenges developers face is that modern applications often need to support multiple versions of a feature—different algorithms or formats that can…
AWSBuilding RAG Systems on AWS: Lessons from Serverless and EC2 Benchmarks
A practical benchmark of RAG ingestion and search performance across AWS Lambda and EC2
BioinformaticsRethinking Bioinformatics
Solving the Challenges of Traditional Pipelines
Strands AgentsPart 2: Agentic Patterns 101 with Loka & Strands-Agents
After exploring the foundations of Agentic Patterns in Part 1, where we looked at more structured workflows like Sequential and Parallel workflows, LLM Routing, and Reflection, this time we are going to push further…
Strands AgentsPart 1: Agentic Patterns 101 with Loka & Strands-Agents
These days, Agents are everywhere in daily conversation and it seems everyone wants one. We read or hear about AI Agents, Agentic AI, Agentic Architectures, and Agentic Patterns almost every day. But what do these terms…
BioinformaticsFlattening the Curve with Nextflow: Building a Scalable and Reproducible Bioinformatic Workflow for MITNANEX
Loka’s participation in the recent nf-core hackathon in Medellín, Colombia, culminated in the successful development of MITNANEX, a pipeline designed to extract and assemble mitochondrial genomes, identify and annotate…
AndroidHow to Transition from XML (Android View) to Jetpack Compose View
Moving from XML to Jetpack Compose helps developers write cleaner, faster and more modern UI code — essential for staying up to date with Android’s best practices.
AWSAWS S3 browser will save you time
Simple interface for data stored in S3
AIPart 4: Benchmarking DeepSeek Cost, Performance and Business Use Cases on AWS
A DeepSeek benchmark on AWS that compares cost, performance, and deployment choices for business workloads using open-weight models.
AWSPart 2: Deploying Distiled DeepSeek-R1 Models on Amazon SageMaker AI
A deployment walkthrough for serving distilled DeepSeek-R1 models with Amazon SageMaker AI, including the AWS setup for open-weight reasoning models.
AWSPart 3: Deploying DeepSeek-R1 Models on AWS- designed Silicon Instances
Engineering notes on deploying DeepSeek-R1 models to AWS-designed silicon instances and serving open-weight reasoning models on AWS.
AWSPart 1: Deploying Distilled DeepSeek-R1 Models on Amazon Bedrock
Deploy distilled DeepSeek-R1 open-weight models through Amazon Bedrock, with a hands-on AWS setup from the Loka engineering team.
DeepSeekDive into DeepSeek
How the open-recipe LLM is transforming GenAI
NetworkingHow HTTP/3.0 and QUIC Solve Head-of-Line Blocking
You probably have heard the history of how HTTP was created and you might even know about the new HTTP/3 version. In this article, I want to focus on some of the lesser-known details of HTTP and why we need a new…
Authors
Bojan Jakimovski- Crhistian Cardona
Petar Kalinovski
Nina Cvetkovska
Alexandre Domingues
Andres Florian
Daniel Sabogal
Ervin Shaqiri- Jelena Pejovic
- Ana Marković
Andreia Pereira- Andrés Sacre
Cecilia Brusquetti- Daniel Pereira
Didier
Dragan Savevski
Federico Rueda
Gabriel Menezes da Silva
Guilherme Ribeiro
Javier Dominguez- Jim Burtoft
João Afonso Pereira- João Correia
Jorge Moura Sampaio
José Ribón
Julián F. Fernández- Juliana Silva
- Louise Ping
Mario Petkoski
Matheus Dias
Nicolás Franco- Paolo Pecis
- Romulo Pagnozzi
- Sara Oquendo
Stefanija Zdraveska- Telmo Felgueira
- Tiago Gonçalves
Zafir Stojanovski