LLM, Data & ML systems

Manos Adamopoulos

Data / AI Engineer

I design, ship and operate Python backends and ML/LLM systems on Google Cloud, owning everything from requirements to CI/CD, deployment and monitoring.

See my projects Download CV

About

Software Engineer with 4 years building production Python services and ML/LLM systems on Google Cloud (Cloud Run, Kubernetes). Two MSc degrees in STEM, both with honors.

I have spent 6+ years working directly with clients in startups, turning real business problems into high-impact technical solutions. Today I build AI Agents you can trust, at the core of GWI’s AI transformation, and I am part of the AI Enablement team defining agentic engineering standards across the organisation.

Education

  • MSc in Software Development & AI

    University of Piraeus · 9.8/10

  • MSc in Civil Engineering (5-year)

    Democritus University of Thrace · 7.04/10

Experience

  1. AI Cloud Engineer @ GWI

    Oct 2024 – Present

    I build and run the AI backends and cloud infrastructure behind GWI’s LLM-driven products, from synthetic survey generation to ML segmentation pipelines, and help set agentic engineering standards across the company.

    Show details

    Lynx – Synthetic Survey Generation Agent (RAG)

    • Architected and shipped a production AI orchestration backend (Python/Django, DRF) for an LLM-driven synthetic survey platform: FSM-based job lifecycle with row-level concurrency control and idempotent retries; parallelized the respondent batch-generation pipeline with a Redis-backed async worker tier and graceful long-job shutdowns, cutting end-to-end runtime by ~94% (77+ PRs, ~400 commits).
    • Deployed and operated cloud-native infrastructure on GKE via custom Helm charts and Terraform/Terragrunt: Workload Identity, KMS-encrypted secrets, Cloud SQL, multi-env (staging/prod) with HPA, PDBs and Prometheus alerting; automated object-storage → BigQuery ingestion feeding the company’s data platform.
    • Built an LLM synthetic-respondent engine: semantic few-shot retrieval over a 483-question, 1,536-dim embedding index, Big5-persona-conditioned generation, largest-remainder demographic quota sampling, and parallel batch execution; created an evaluation harness benchmarking synthetic vs. real respondents across 5 metric families.

    Segmentation Integration Automation (SIA) – ML pipeline predicting custom audience segments on GWI’s Core dataset

    • Engineered and productionized an event-driven ML post-processing microservice on GCP (Cloud Run, Pub/Sub, Docker, Terraform) that rebalances weighted audience distributions for tens of thousands of respondents per run, guaranteeing convergence within 0.5% of market targets via a deterministic, parallelized iterative optimizer with early stopping.
    • Designed a novel LLM-agent optimizer (GPT-4o function-calling loop, 3 custom domain tools) that autonomously measures distribution divergence and reweights segment assignments; extended the engine to multilabel segmentation and achieved 100% ground-truth respondent retention, fixing a production-critical reassignment defect.

    Platform and company-wide initiatives

    • Led adoption of engineering standards for the department as DevOps Champion, driving AI agentic engineering harness adoption via the AI Enablement team and migrating department infrastructure into GWI’s central Terraform repo with resource labelling that enabled per-product cost attribution.
    • Drive AI literacy across the organisation: speak at public discussions and run an AI News Forum that keeps teams informed on developments and encourages adoption.

  2. Data Engineer @ GWI

    Oct 2023 – Oct 2024

    I modernised GWI’s data processing and connected its ML pipeline to product APIs, replacing slow legacy workflows with versioned, automated ones.

    Show details

    • Integrated the ML pipeline with the product team’s Segmentation API over authenticated REST calls, decoupling GCS file I/O from the workflow engine.
    • Replaced legacy SPSS processing with a BigQuery ETL pipeline, consolidating 3 data sources into one version-controlled transformation layer. The pipeline cut its runtime from 12 hours to 1 hour (~90% reduction).

  3. Solutions Engineer @ Linked Business

    Mar 2021 – Oct 2023

    Market and credit intelligence (KYC) platform covering 800K+ Greek companies; clients include systemic banks and credit-servicing firms.

    I was the technical link between the platform and its banking clients: scoping integrations, fixing recurring pain points, and turning client feedback into shipped features.

    Show details

    • Owned technical relationships with systemic banks and credit servicers, leading integration scoping, security reviews, and go-live of KYC onboarding and credit-check workflows; turned client pain points into prioritised roadmap fixes.
    • Cut authentication tickets 40% in one quarter via root-cause analysis of recurring login failures; shipped the fix with Playwright regression tests in CI/CD so the failure class cannot silently return.
    • Drove 50% API usage growth in 6 months: turned client interviews into a prioritised roadmap item, then scoped and delivered a bulk Excel export endpoint.

  4. Technical Project Manager @ Sociality co-op

    Dec 2019 – Mar 2021

    I led a small engineering team building web platforms for a cooperative, from planning through delivery.

    Show details

    • Led a team of 6 developers to deliver web platforms, including a unified CMS, member portal and ticketing system (TypeScript, React, Node.js, PostgreSQL), driving a 40% increase in ticket sales within one quarter.

Personal projects

J

Job Matcher

A full-stack AI platform where job seekers describe roles in natural language, search listings from extracted preferences, and get concise explanations of why results fit.

  • Sequential multi-agent backend (preference extraction and match advisor) with evidence-backed prompts from vector retrieval over API docs (text-embedding-3-small, ChromaDB).
  • Failure-tolerant RAG integration: lazy service init, and it keeps working without context on errors.
  • React/TypeScript UI, FastAPI services, Docker images and Kubernetes deployment.
  • GitHub Actions CI builds backend and frontend images on every push/PR to main.
  • Python
  • FastAPI
  • React
  • TypeScript
  • OpenAI
  • ChromaDB
  • Docker
  • Kubernetes
Code ↗
S

SkyScribe-CLI

A multi-agent Python CLI that turns live forecast data into short, actionable weather briefs (temperature ranges, rain windows, wind, practical tips) for travelers and daily planning.

  • Uses the Open-Meteo APIs for live forecast data.
  • Fine-tuned TinyLlama with LoRA adapters (Transformers + PEFT): train and ship small adapters instead of full weights.
  • Supervised pairs with prompt-token label masking, optional 4-bit QLoRA for GPU-efficient training, and adapter hot-loading at inference via env config.
  • Python
  • Typer
  • httpx
  • Hugging Face
  • PEFT
  • LoRA
Code ↗

Skills

Languages

  • Python
  • SQL
  • TypeScript
  • JavaScript
  • HTML/CSS

Backend & APIs

  • FastAPI
  • Django REST Framework
  • REST API design
  • Pub/Sub
  • PostgreSQL
  • Redis

Cloud & DevOps

  • GCP (Cloud Run, GKE, BigQuery, GCS)
  • Docker
  • Kubernetes / Helm
  • Argo Workflows
  • Terraform
  • GitHub Actions

ML / LLM

  • RAG
  • Embeddings
  • ChromaDB
  • LightGBM
  • scikit-learn
  • SHAP
  • Hugging Face / PEFT

Reliability & security

  • Secret Manager / SOPS
  • Sentry
  • Structured logging
  • pytest
  • Playwright

Frontend

  • React
  • Tailwind

Contact

Let’s build something.

I’m always happy to talk about AI systems, data engineering, or interesting problems. The inbox is open.

Say hello