( Portfolio — 2026 )

I build AI systems that generate, evaluate, act, remember, and improve.

I design and ship custom AI systems that combine models, tools, data, feedback loops, and automation. My work spans agentic systems, synthetic data, local deployment, evaluator loops, and self-improving pipelines built for real-world use.

Core System Pattern
A reusable closed loop for systems that get better over time.
01  Goal  →  02  Plan  →  03  Build Context 04  Execute  →  05  Evaluate  →  06  Memory
Self-
Improving
Loop
1
Goal
2
Plan
3
Build Context
4
Execute
5
Evaluate
6
Memory / Improve
01

How to describe my work

I build agentic AI systems that plan, act, and learn.

I design custom AI systems with tools, memory, evaluators, and self-improving workflows.

I architect production-ready AI systems that integrate models, tools, data, and feedback loops to deliver measurable outcomes.

Best fit for
Product teams building custom AI features
Operations teams automating workflows
Companies needing local / private AI
Teams that want systems beyond simple chatbots
Typical outcome Better quality Less manual work Lower cost Improves over time
02

Core AI Capabilities

The main technical capabilities I use to design, build, and ship intelligent systems.

01

Agentic Systems Engineering

Design custom agents that plan, act, and adapt across tools and workflows.

custom AI botsmulti-step planningpersistent memoryhuman-in-the-loopmulti-agent coordination
02

Tool-Native Automation

Build systems that execute work through APIs, files, dashboards, CRMs, and internal tools.

tool chainingAPI integrationretries / fallbacksaccess controlevent-driven
03

Context Engineering

Assemble the right context for the right task from memory, tools, docs, and state.

retrieval / rerankingsummarizationconversation statelong-term memoryrelevance validation
04

Synthetic Data / AutoData

Generate, expand, and curate training, testing, and evaluation data.

taxonomy growthsynthetic samplesaugmentationdedup / labelingprivacy-preserving
05

VLM / LLM Feedback Loops

Use models as evaluators and critics to score outputs and drive improvement.

visual QAoutput critiquerubric scoringiterative refinementpreference ranking
06

Local / Edge AI Deployment

Run models privately and efficiently on local, browser, or edge infrastructure.

ONNX runtimesquantizationWebGPU / WASMoffline workflowssecure inference
07

Model Improvement Pipelines

Continuously test, fine-tune, and improve systems with data and feedback.

golden setsautomated evalLoRA / adaptersA/B testingregression tracking
08

Workflow Automation & Delivery

Turn AI systems into reliable, deployable, documented workflows.

dashboardsrunbooks / docsQA checklistshandover / trainingproduction support
03

System Blueprints

Reusable AI system architecture patterns I can design and implement end-to-end.

P1

Tool-Native Business Agent

01Goal
02Planner
03Context Builder
04Tool / API Calls
05Approval / Action
06Logs / Memory

Value  Automates end-to-end business workflows with tools, approvals, and learning memory.

P2

AutoData / Synthetic Dataset Engine

01Seed Concepts
02Generate Variants
03Render / Simulate
04Evaluate / Filter
05Caption / Label
06Archive / Taxonomy

Value  Scalable synthetic data generation for training, testing, and rapid domain coverage.

P3

Self-Improving Model Loop

01Baseline Model
02Generate Outputs
03Critique (VLM / Rules)
04Score
05Refine Data / Prompts
06Re-test ↺

Value  A continuous improvement loop that refines data and models using automated critique and scoring.

P4

Local Agentic Research Loop

01Objective
02Generate / Explore
03Evaluate
04Compare
05Select
06Archive Learnings

Value  Local closed-loop experimentation for private, iterative AI research and system improvement.

Reusable ingredients Planners Evaluators Memory Scoring Tool Execution Human Approval Deployment Monitoring
04

What I Can Build

01

Custom AI Bots

Task-focused bots for support, research, operations, and specialist workflows.

02

Internal Copilots

Knowledge-aware copilots that help teams find, create, and act on information.

03

Autonomous Automation Agents

Agents that plan, execute, and adapt across systems.

04

Data Generation Engines

Synthetic data and scenario generation pipelines.

05

Taxonomy Expansion Systems

Systems that evolve and maintain classes, attributes, and labels.

06

Evaluator / Critic Loops

LLM / VLM evaluators that score, critique, and improve outputs.

07

Local AI Workflows

Private local-first AI systems with strong control.

08

Browser / ONNX Inference Apps

Fast browser inference using ONNX / WebAssembly.

09

Analytics Dashboards

Dashboards surfacing metrics, trends, and model behavior.

10

QA / Evaluation Pipelines

Automated dataset / model evaluation and regression tracking.

05

Tech Stack & Delivery

Typical tech stack
ModelsOpenAI (GPT-4o, GPT-4), LLaMA 3.x, local models (Mistral, Phi, Qwen), VLMs.
OrchestrationCrewAI / LangGraph, custom agent loops, workflow engines.
DeploymentONNX Runtime, browser inference, local / cloud (Docker, VM, K8s), APIs.
Data & ToolsSQL / databases, files / object storage, dashboards, CRMs / apps / webhooks.
EvaluationVLM judges, automated metrics, scoring / rubrics, human-in-the-loop.
Delivery & working style
01

Discover

Understand goals and constraints.

02

Prototype

Validate ideas fast.

03

Integrate

Connect models, data, tools.

04

Evaluate

Test quality and safety.

05

Iterate

Refine prompts and logic.

06

Document

Runbooks and usage docs.

07

Train

Enable your team.

08

Handover / Support

Smooth handover and support.