available for Q4 projects ots
what
i
do
i build the full path around a generative model: the data it learns from, the workflows that run it, the checks that keep it consistent, and the app a team uses to run it.
(01)
GENERATIVE AI PIPELINES
from dataset to production: the training, the workflows and the checks around a generative model.
COMFYUI WORKFLOWS
Custom LORA TRAINING
DATASET & CAPTIONS
VISION QA
MULTI-MODEL PIPELINES
(02)
AI PRODUCT / ENGINEERING
the app around the model, so a team can use it without an engineer.
FASTAPI BACKENDS
REACT + TYPESCRIPT
SELF-SERVE PLATFORMS
ML SERVICE INTEGRATION
DOCKER + CI/CD
(03)
AGENTS & AUTOMATION
agent systems and workflows that take over repeated work, with a review step before anything ships.
MULTI-AGENT HARNESSES
N8N WORKFLOWS
RAG SYSTEMS
CLAUDE + OPENAI APIS
AUTOMATED REPORTING
(04)
WEB & E-COMMERCE
websites and stores built to be measured: analytics first, then the fixes.
SHOPIFY BUILDS
WORDPRESS / WOOCOMMERCE
FRAMER SITES
GA4 + BIGQUERY
CONVERSION WORK
How i
build
Preparation
reference images, captions and a controlled vocabulary, so the model learns the right identity.
Train
LoRA training on local and cloud GPUs, with each checkpoint audited before it's used.
Build
versioned ComfyUI workflows, with a vision QA agent scoring every output.
Hand over
a self-serve app so the team runs it without me.
core
capabilities
GENERATIVE
FLUX LoRA Training
Caption Validation
ComfyUI Workflows
Dataset Engineering
Image-to-Video
ENGINEERING
React + TypeScript
Docker + CI/CD
Python
PostgreSQL
Azure / AWS / GCP
AGENTS
LangGraph
RAG
Multi-Agent Systems
Hermes Agents
EVALS
Model Evaluation
Rubric Scoring
Vision QA Agents
Versioned Workflows
FAQ
Question
User
do you only work with ComfyUI?
1
No. ComfyUI and Weavy (Figma Weave) are where the image and video workflows run, but the system around them is Python, FastAPI, React and a vision QA agent. i pick the tools after i've seen the problem.
Question
User
can i see Client's work?
1
not publicly, it's under NDA. the architecture and decisions are on a password-protected page, and i can give access to people who ask.
Question
User
what does your process look like?
1
four steps. prepare the data, train the model, build the workflows with checks on every output, and hand it over so the team can run it without me. my Bernadette work followed exactly this.
Question
User
do you build RAG systems?
1
Yes. i've built a RAG workflow in n8n for a client. the details are under NDA, so it sits behind the same password-protected page.