Available for Technical Collaboration Yerevan, Armenia (UTC+4)

AI Systems Architect

Designing reliable ML, LLM, and agentic workflows—from data pipelines to deployment.

Co-Founder & CTO at Ancanot and AI Systems Architect. I build high-throughput feature pipelines in Polars and Dagster, robust vision models under severe data constraints, and containerized microservices—extending practical engineering foundations into reliable LLM and agentic workflow architectures grounded in verifiable code and explicit architectural trade-offs.

Polars + Dagster Lazy Join DAG & Protocol Inversion
93% CV Accuracy Sidewalk Hazard Detection (VGG Systems)
PostgreSQL + OCR Relational Biomarker Caching Pipeline
AUA CS Senior Algorithms & Distributed Systems
Portrait of Zaven Avagyan

Zaven Avagyan

AI Systems Architect · CTO @ Ancanot
Yerevan, Armenia · American Univ. of Armenia
AI Systems Architecture Designs modular ML systems across data pipelines, model inference, and application services, with clear component boundaries and practical deployment constraints.
Agentic & LLM Application Design Builds LLM-enabled workflows and prototypes connecting language models with structured context, document processing, and application logic.
ML Platform & Pipeline Engineering Builds reproducible feature pipelines and model-serving services, including lazy Polars backfills, Dagster orchestration, and containerized transformer inference.
Applied AI Technical Leadership Co-founded Ancanot as CTO and led ML engineering, technical team direction, and AI curriculum development across prior roles.

Engineered Systems & Verifiable Implementations

Spotlighting top flagship engineering implementations first, followed by production microservices and systems proposals. Every project is backed by authentic repository code, verifiable pytest validation suites, or documented industry implementations. Zero unbacked claims or decorative dashboards.

Interactive Dataflow & Technical Choices

Inspect the stage-by-stage dataflow, memory policies, and architectural boundaries of implemented pipelines and microservices. Click or focus any stage to inspect the technical rationale.

Stage 01: Source Ingestion — Architectural Detail

Lazily scans event logs and temporal partitions into Polars LazyFrame without eager RAM materialization.

Core Engineering & Applied ML Capabilities

Organized strictly by verified domain competence grounded in repository implementations, industry roles, and verified academic training. No arbitrary percentage bars or decorative charts.

Machine Learning & Deep Learning

Statistical modeling, neural architectures, and optimization applied to concrete problems.

PyTorch & TensorFlow Deep learning models across CV and NLP production pipelines
Transformers & Hugging Face DistilBERT inference microservices, GPT integration, LLM fine-tuning
Computer Vision (YOLO, ResNet) Sidewalk hazard detection with 93% accuracy under training data constraints
Scikit-Learn, XGBoost, LightGBM Tabular feature modeling, cross-validation, hyperparameter tuning
Model Evaluation & Metrics Cross-validation, ROC-AUC, Precision, Recall, F1-Score, preventing data leakage

Data Systems & Pipeline Engineering

High-throughput, memory-conscious data manipulation and distributed computing.

Polars (Lazy Execution) Engineered feature backfill engine with single-collect memory optimization
Dagster Orchestration Decoupled asset dependencies using Python Protocols and date-partitioned runs
PostgreSQL & Relational SQL Relational persistence, biomarker normalization, and indexing in Medical Assistant
Pandas & NumPy 4+ years of data handling, vectorized computation, and numerical analysis
Distributed Computing (Spark / Hadoop) Distributed compute architectures and batch processing patterns
Document Extraction (OCR) Pytesseract, PDFplumber, PyMuPDF for automated document ingestion

Production Engineering & MLOps

Containerization, cluster configuration, API design, and system reliability.

Docker & Containerization Multi-stage builds, model caching, minimal image packaging
Kubernetes (K8s) Declarative deployments, ClusterIP services, resource requests/limits
FastAPI & REST APIs Asynchronous microservices with Pydantic request validation
Cloud Platforms (AWS & GCP) Cloud computing, storage, container deployment
Automated Testing (Pytest) Unit testing temporal parity, resource API validation

Technical Leadership & Credentials

Steering engineering direction, teaching technical courses, and competitive modeling.

Co-Founder & CTO (Ancanot) Directing technology strategy, distributed system architecture, and product execution
IT Team Lead (SellStorm Group) Conducted architecture & code reviews, mentored engineering teams
AI Course Instructor Developed and taught an intensive course covering ML theory and practical applications
CaseKey 2024 Finalist Verified on Certopus (Credential ID: 81d1477fd40849efadea4fa0bf134c6c)
Datathon 2023 Winner First place in competitive predictive modeling competition

Engineering Leadership & Industry Experience

Accurately distinguishing executive technical leadership, senior ML engineering and pedagogy, and applied deep learning engineering.

Co-Founder & CTO

Jan 2026 — Present
Ancanot · Yerevan, Armenia Executive Technical Leadership

Leading the technical vision and development of a social communication platform focusing on communities, voice calls, and meaningful interaction. Overseeing technology strategy, system architecture, product development, and engineering direction to build scalable and reliable communication solutions.

  • Directing overall technology strategy, distributed system architecture, and product engineering.
  • Architecting scalable, resilient backends designed for low-latency communication and privacy.
  • Steering engineering roadmap, system reliability, and cross-functional product direction.

Senior ML Engineer & IT Team Lead

May 2025 — May 2026
SellStorm Group · Yerevan, Armenia Senior Engineering & Pedagogy

Led end-to-end development and deployment of ML systems, designing scalable pipelines and ensuring production-grade performance. In addition, developed and taught an AI curriculum covering theoretical foundations and practical applications, and served as IT Team Lead conducting architecture and code reviews.

  • Designed scalable machine learning pipelines ensuring production-grade serving and throughput.
  • Served as IT Team Lead: conducted system architecture and code reviews, mentoring engineering teams.
  • Developed and taught an intensive AI course covering foundational statistical theory and practical applications.

Machine Learning Engineer

May 2022 — Dec 2024
VGG Systems · Yerevan, Armenia Machine Learning Engineering

Developed and optimized deep learning models for computer vision and NLP applications, including object detection and image classification systems using CNNs, YOLO, and transfer learning (ResNet), achieving 93% accuracy on sidewalk hazard detection with limited data. Also fine-tuned an LLM for an Armenian conversational chatbot.

  • Engineered sidewalk hazard detection with CNNs, YOLO, and ResNet transfer learning, achieving 93% accuracy under severe training data scarcity.
  • Applied advanced training strategies: synthetic data augmentation, cosine learning rate scheduling, and early stopping.
  • Fine-tuned conversational LLMs for Armenian language processing, accounting for dialectal nuances and morphological complexity.

Education, Honors & Verifiable Credentials

Formal university education in computer science, specialized academic programs in machine learning, and third-party verified competition credentials.

Formal Education

Bachelor of Science in Computer Science

Present / Ongoing
American University of Armenia (AUA) · Yerevan

Core focus in algorithms, distributed systems, software engineering, and applied mathematics.

Machine Learning Academic Program

Completed March 2026
Business Development Group (BDG) · Yerevan

Advanced curriculum covering deep learning architectures, statistical learning theory, and MLOps deployment patterns.

International High School Diploma

Graduated Jan 2020
“Shirakatsy Lyceum” International Educational Complex · Yerevan

Specialized foundations in advanced mathematics, physics, and bilingual scientific instruction.

Honors & Verified Certifications

CaseKey Case Competition Finalist

2024
Nation 2.0 Educational NGO

Advanced through all competitive rounds to the finals, demonstrating analytical modeling and structured strategic problem solving.

Verify Credential on Certopus (ID: 81d1477fd40849efadea4fa0bf134c6c)

Datathon Winner

2023
National Data Science Competition

First-place prize for outstanding machine learning predictive modeling and rapid analytical problem solving.

Aspire Leaders Program

2025
Aspire Institute

Selected for prestigious global leadership program focusing on social impact and technology innovation.

CaseKey Academy Completion

2024
CaseKey Academy

Rigorous certification in structured strategic problem solving and analytical evaluation.

Language Proficiencies
Armenian Native / First Language
English C2 Proficient
Russian C2 Proficient

Let's build serious AI systems architectures together.

Whether you are evaluating architectural direction for ML pipelines, designing reliable LLM and agentic workflows, discussing engineering leadership, or exploring technical collaboration—feel free to reach out directly.

Based in Yerevan, Armenia (UTC+4) · Open to high-impact distributed collaborations