09/12/2026 updated

**** ******** ****
100 % available

Senior AI Engineer | LLM Agents, GenAI & Document AI | AWS Certified Architect | MLOps | 8+ yrs

Bilje, Croatia
Only remote
M.S. Computer Science (Data Science) and B.S. Computer Science, FERIT Osijek, 3 peer-reviewed papers
Bilje, Croatia
Only remote
M.S. Computer Science (Data Science) and B.S. Computer Science, FERIT Osijek, 3 peer-reviewed papers

Profile attachments

Brando_Koch_CV.pdf

About me

Expert in AI with 8+ years of experience, specialized in ML, MLOps, and AWS, in both development and research settings. I help startups and enterprises design and implement end-to-end AI systems, from data collection, through model development, to deployment.

Data MeshBedrockSageMakerLangChainArtificial IntelligenceAmazon Web ServicesOptical Character Recognition (OCR)Computer VisionInformation RetrievalPython (Programming Language)PostgreSQLMachine LearningNatural Language ProcessingTensorFlowSpeech RecognitionChatbotsPyTorchLarge Language ModelsDeep LearningGenerative AIFastAPIAWS Data AnalyticsMachine Learning OperationsTerraformDocker
Machine Learning & Deep Learning
Expertise in designing, training, and deploying machine learning and deep learning models, including transformer-based neural networks, NLP, conversational AI, and speech recognition systems. Hands-on experience with PyTorch, TensorFlow, Keras, HuggingFace, and SageMaker across multiple production environments.

Generative AI & LLM Engineering
Extensive experience in building and deploying generative AI systems, including LLM training, fine-tuning, and agentic AI architectures. Proficiency with LangChain, LangGraph, OpenAI, Gemini, Anthropic, AWS Bedrock, and DeepAgents for end-to-end AI product development.

AWS & MLOps
Deep expertise in AWS cloud infrastructure and MLOps, including system architecture, scalability, and reliability. Certified AWS Solutions Architect with hands-on experience in SageMaker, EC2, ECS, S3, RDS, Lambda, Lake Formation, Terraform, and end-to-end infrastructure design.

Natural Language Processing (NLP)
Solid background in NLP tasks including semantic question answering, sentiment analysis, text classification, and speech-to-text. Application of NLP in production systems for audio quality control, call center analytics, and document understanding.

Computer Vision
Experience in computer vision applications including OCR using Vision Language Models (VLMs), parking occupancy detection, and neural style transfer. Development of novel approaches for extracting and structuring visual information from documents.

Data Engineering & Data Infrastructure
Design and deployment of enterprise-grade AWS Data Mesh solutions, multi-region data platforms, and data pipelines. Proficiency with Lake Formation, Glue, RDS, SNS, and Terraform for scalable and governed data infrastructure.

Speech & Audio AI
Development of end-to-end speech recognition systems, speaker diarization, voice activity detection, text-to-speech models, and audio data sorting. Experience with Whisper architecture, Nvidia Nemo, and custom audio preprocessing pipelines.

AI System Architecture & Technical Leadership
Ownership of core AI platform architecture from system design to engineering execution, including technical estimation, roadmapping, and leading engineering teams. Experience advising CEOs on AI strategy and managing Series A startup environments.

Python & Software Engineering
Proficiency in Python as the primary programming language for AI and ML development. Experience with Docker, FastAPI, PyTorch, Streamlit, and various ML frameworks for building and packaging production-ready AI solutions.

Experience

Professional experience5 - 10 years
Freelancer since2022

Languages

GermanBasic knowledgeEnglishFluentCroatianNative speaker

Project history

CEO

intelligentia.ai, LLC
Owner of Intelligentia, an AI research and development company. Offering experience across the full AI lifecycle: from leading AI research and development, to designing AI products, systems, and infrastructure, to advising CEOs on AI strategy.

AI Advisor

FileAI, Pekin, Singapore
Advised FileAI on agentic AI direction and shipped SuperAgent, the company's customer-facing AI assistant that searches a company's own documents, reasons over what it finds, answers with grounded citations, and generates polished PDF reports.

Lead AI Engineer

FileAI, Pekin, Singapore
Led AI strategy and development for a Singapore startup building world-class data extraction and data entry automation technology for files. As a direct report to the CEO, owned the company's core AI platform end-to-end, from system architecture across AI and AWS, through technical estimation and roadmapping, to leading and overseeing engineering execution. Series A raised 14M USD.

Lead AI Engineer, GenAI Document Processing (OCR and form filling)

FileAI, Singapore

Internet & IT

Helped FileAI, an AI-native document processing platform handling close to half a billion documents per year across US, Asia, EU and India, adopt Generative AI for its two core operations: OCR and form filling. Developed and trained LLM and VLM models coupled with text embedding models that substantially increased OCR and form-filling accuracy on complex documents spanning foreign languages and alphabets (e.g. Thai, Japanese) and difficult layouts (long tables, checkboxes, stamps). Enabled semantic representations of documents containing visual elements such as images and graphs. Built the supporting AWS infrastructure (EC2, ECS, S3, RDS, CodePipeline, Lambda, Bedrock, SageMaker) with autoscaling, plus a monitoring solution giving live visibility into every AI model call. The solution serves FileAI's production workload. Client quote (CEO): deep knowledge of the AWS cloud platform and a progressive stance on generative AI. Stack: LangGraph, LangChain, OpenAI, Gemini, Anthropic, AWS Bedrock, SageMaker, vLLM, Terraform, PostgreSQL, LangSmith.

AWS/MLOps Engineer, Private Amazon SageMaker Studio Domain in Terraform

Confidential client (data science team)
Set up an Amazon SageMaker Studio domain in Terraform for an organization whose data scientists needed a browser IDE while no data, notebook or model artifact was allowed to leave the AWS account. VPC-only mode with no internet route, relying on interface and gateway endpoints for the SageMaker API, runtime and S3; domain security group created and attached explicitly; ECR pull/push restricted to trusted accounts. Authentication through IAM Identity Center with one execution role per user profile so each user's permissions are explicit. Replaced the auto-created EFS with a customer-managed, KMS-encrypted EFS with mount targets per subnet and per-user EFS access points (own POSIX uid/gid, mode 700) so users cannot read each other's files; an EventBridge rule on the CreateUserProfile CloudTrail event triggers a Lambda that creates each new user's directory automatically. Guardrails (domain, network, roles, storage, SSO assignment) live in code, while spaces are self-service; idle shutdown enabled to control compute cost. Stack: Amazon SageMaker Studio, Terraform, VPC endpoints, IAM Identity Center, EFS, KMS, EventBridge, Lambda.

AI Engineer, AI customer-support chatbot on AWS (RAG over website content)

Tellofy

Internet & IT

Partnered with Tellofy, a customer-experience platform, to build an AI-powered website support chatbot that indexes website content and internal knowledge bases and answers customer questions directly in a chat interface. Delivered in phases: a locally deployed proof of concept to validate feasibility and performance, then deployment on AWS integrated with a Streamlit demonstration website, then integration of the API backend into the central Tellofy dashboard so Tellofy's clients can configure and customize the chatbot for their own websites. AWS services: EC2, S3, Lambda, Bedrock, CodePipeline, Application Load Balancer. Client statement (Founder and CEO): the solution automated a significant portion of customer support on their clients' websites. Stack: RAG, AWS Bedrock, Python, Streamlit, AWS.

AI Consultant, Generative AI enablement on AWS Bedrock (RAG pilot)

ComeToTheClouds (CTTC), AWS consultancy

Internet & IT

Helped ComeToTheClouds, an AWS consultancy, add Generative AI capabilities to its AWS environment with security, privacy, cost-efficiency and scalability as priorities. Ran a strategic assessment that identified AWS Bedrock as the best-ROI path, then delivered a pilot: configured Bedrock with Cohere Embed 3 Multilingual and Anthropic Claude 3.5 Sonnet using cross-region inference for throughput, deployed Pinecone as the vector database via AWS Marketplace, and built a FastAPI plus LangChain Retrieval-Augmented Generation API for document ingestion, embedding, retrieval and LLM question answering, deployed on ECS Fargate and orchestrated with Terraform. Client statement (CEO): thanks to the engagement they can now leverage Generative AI on AWS. Stack: AWS Bedrock, Pinecone, FastAPI, LangChain, ECS Fargate, Terraform.

Senior AWS/MLOps Engineer

Come to the Clouds d.o.o
Working on AWS and Data Infrastructure solutions. Ensuring scalability and reliability. Performing end-to-end infrastructure system design on AWS from networking to compute and storage.

Senior AWS/MLOps Engineer, Enterprise AWS Data Mesh (Lake Formation, Terraform)

Global reinsurance leader (via Come to the Clouds d.o.o.)

Banking & Financial Services

Core engineer on a multi-account enterprise data mesh on AWS for a leading global reinsurer, delivered while at Come to the Clouds. Problem: secure data access required multi-week approval chains and manual permission management across isolated AWS accounts, blocking the company's AI strategy. Designed and implemented a Lake Formation based mesh with three account types: data producer accounts (domain-owned S3 data lakes), a central governance account (shared Glue Data Catalog, Glue crawlers, canonical tag vocabulary) and data consumer accounts (Athena and similar engines). Chose Lake Formation tag-based access control (LF-TBAC) so grants scale as n(P) + n(R) instead of n(P) x n(R), with sensitivity, domain and tier tag dimensions and a self-serve onboarding and sharing mechanism. Defined the producer, governance and consumer roles and permission flow. Everything (account onboarding, roles, Lake Formation config) delivered as pure Terraform, multi-region and multi-environment, giving an auditable record by construction. Outcome reported by the customer: data-access provisioning went from weeks to minutes, and onboarding new producers or consumers adds negligible central work. Stack: AWS Lake Formation, Glue, Athena, S3, IAM, Terraform, Terragrunt, GitHub Actions.

AI/MLOps Engineer, PrivASR: private high-throughput speech-to-text engine on AWS

Intelligentia (privatesynapse.ai product)

Internet & IT

Designed and built a private, self-hosted speech-to-text engine for teams that need to transcribe large volumes of audio without sending recordings to a third party. Pipeline: pyannote voice activity detection, NeMo clustering diarization (TitaNet-L speaker embeddings) and faster-whisper large-v3 via CTranslate2 on GPU, merged by maximum temporal overlap so every line carries a speaker and timestamp. Whisper hallucination over silence suppressed with the Silero VAD filter; beam search decoding; configurable float16/int8 compute. Infrastructure: two GPU worker types on ECS fed by SQS, with a scale-from-zero design (Lambda queue poller sets desired count, then target tracking on queue depth, ECS capacity provider grows a zero-minimum Auto Scaling Group) so idle cost is zero. No API by design: jobs are S3 plus SQS messages, results land in S3 with per-segment text, speaker and per-stage timings; access control is plain S3/SQS permissions. Per-stage real-time-factor metrics in CloudWatch, errors in Sentry, Streamlit reference app on Fargate. Result: roughly 10x faster than realtime at under 0.05 USD per audio hour, provisioned with Terraform and deployable into a client's AWS account in essentially one step. Stack: Python, PyTorch, Whisper, NVIDIA NeMo, pyannote, Docker, ECS, SQS, Lambda, S3, CloudWatch, Terraform.

Senior AI Engineer

OnscriptAI LLC
Led AI for an audio analysis product company. Drove solution design, LLM training and fine-tuning, and advised on data collection and product integration.

Senior Machine Learning Engineer

Digital Media Solutions Inc.
Worked in the area of ML, NLP, and Conversational AI. Focused on training and deployment of transformer-based neural network models for audio quality control analysis and insights, utilizing automatic speech recognition, semantic question answering, and sentiment analysis.

Machine Learning Engineer

Yodel Technology LLC
In charge of developing Conversational AI solutions. Responsible for all stages, from data collection to model deployment. Acquired by Digital Media Solutions.

Machine Learning Engineer, Natural-prosody text-to-speech voice for customer support

Yodel Technology LLC (acquired by Digital Media Solutions)

Internet & IT

Developed and deployed a customizable, natural-sounding text-to-speech model based on Tacotron2 for an end-to-end customer support solution at Yodel Technology (later acquired by Digital Media Solutions). Researched and tested existing TTS methods and models, designed the data collection process and labeling guide on the Prodigy annotation platform including a precise recording manual for the voice talent (punctuation-to-prosody rules, emphasis, stable pacing), performed voice cloning by fine-tuning Tacotron2 on the proprietary dataset, and packaged the solution for production. The model learned natural sentence transitions down to the in-breath before speaking, making the synthesized voice hard to distinguish from a human. Stack: Python, PyTorch, Tacotron2, Prodigy, Docker, Triton Inference Server, AWS.

Machine Learning Internship

Atos Convergence Creators d.o.o
Developed a Trigger Word Detection Neural Network Model for a hands-free checklist-filling solution in challenging environments. Audio data preprocessing and feature engineering.

Certificates

AWS Certified Solutions Architect

Amazon Web Services

2024

AWS Certified Data Analytics - Specialty

Amazon Web Services

2024

AWS Certified Database - Specialty

Amazon Web Services

2024

Natural Language Processing Specialization

deeplearning.ai

2021

Deep Learning Specialization

deeplearning.ai

2019

Machine Learning

Stanford University (Coursera)

2019


Contact form

Log in to get in touch

You need to be logged in to use the contact form.

Sign upLog in