AI INFRASTRUCTURE PORTFOLIO

Building Intelligent Retrieval Systems & AI Infrastructure

Selected product work across vector databases, semantic search, MCP tooling, AI agents, Playwright automation, and enterprise delivery.

View Architecture GitHub Profile Explore Projects

RAG systems MCP orchestration Agent memory Enterprise delivery

Semantic topologylive graph

AI CoreEmbedding + retrieval + agent routing

Retrievaltop_k

Memorycontext

Testingplaywright

Toolsmcp

query pulse

Production surfaces0

Web app, browser extension, VS Code extension, and container image

Core domains0

Vector DB, RAG, MCP, agents, search, automation, and workflows

Delivery paths0

GitHub Pages, GHCR container publishing, and extension packaging

Build reliability0.00%

Validated across local builds and distribution targets

Stack depth0

Next.js, React, TypeScript, Tailwind, Framer Motion, Three.js

SELECTED WORK

Real AI engineering products from this workspace

Instead of demo placeholders, this section is grounded in the shipping surfaces and delivery paths already present in the repository.

Project 01

Weaviate UI

A browser-based workbench for inspecting schema, browsing objects, preparing inserts, and running GraphQL queries against Weaviate.

Core product experience for vector database operations.

ReactTypeScriptViteWeaviate

View source Open docs

Project 02

VS Code Extension

A companion editor extension that syncs the same workbench into a webview so teams can work inside VS Code.

Distribution into the developer workflow without changing the UI model.

VS CodeWebviewPackagingSync pipeline

View source Open docs

Project 03

Chrome Extension

An extension build path that packages the app for browser access and local runtime testing.

Expanded access through unpacked and packaged extension delivery.

Chrome ExtensionsBuild syncStatic assets

View source Open docs

Project 04

GitHub Pages + Container Delivery

Static hosting and container publishing workflows that keep the app free to access and easy to ship.

Zero-cost public hosting and reproducible deployment output.

GitHub ActionsPagesGHCRContainerfile

View source Open docs

EXPERIENCE

Engineering patterns behind the portfolio

This is a portfolio-ready view of the systems work already visible in the repo: product UI, tooling, automation, distribution, and native packaging.

AI Retrieval Interfaces

Designed interfaces for vector search, schema-aware exploration, and production query workflows.

Agent Tooling & MCP

Structured tool-routing and context-management patterns for agent-led workflows.

Automation Systems

Built Playwright-friendly verification flows and long-running workflow orchestration patterns.

Packaging & Delivery

Shipped the same product across web, extension, container, and static hosting targets.

VECTOR DATABASE SHOWCASE

Premium retrieval foundations for the product stack

A curated view of the similarity engines that power vector search, RAG pipelines, semantic retrieval, and the Weaviate UI workflow.

Selected retrieval engine

Weaviate

Similarity search pulse

vector query

Composable modules for RAG, agents, and structured knowledge graphs.

Strong schema-centric scaling for enterprise use cases

RAG ARCHITECTURE

How the retrieval system is organized

The visual pipeline maps directly to the semantic retrieval, agent context, and response orchestration patterns used across the workbench and related tooling.

01User QueryNatural language request enters the system.

02Embedding ModelEmbedding model encodes semantic intent.

03Vector DatabaseVector database stores and retrieves context.

04RetrieverRetriever selects the best supporting chunks.

05LLM Context InjectionContext is injected into the model window.

06AI ResponseThe grounded answer is streamed back.

MCP TOOLING

Tooling and context management

This panel reflects the orchestration layer behind agent-driven apps: routing, memory, and command execution patterns seen across the portfolio stack.

[project] weaviate-ui browser workbench ships schema + query tooling
[project] vscode extension packages the same UI inside an editor surface
[project] chrome extension syncs dist assets for unpacked deployment
[release] github actions publish pages, container image, and checks
▍

MCP Workbench

Unified control plane for tools, prompts, and agent context.

command.ready

Tool orchestration

Declarative routing for model-driven execution paths.

command.ready

Context management

Persistent memory windows with scoped retrieval and recall.

command.ready

Agent memory systems

Long-term memory patterns with controlled refresh cycles.

command.ready

PLAYWRIGHT AUTOMATION

Automation and validation workflow

The same project family includes browser, extension, and desktop delivery, so the automation story focuses on repeatable release and validation patterns.

1Launch

2Authenticate

3Validate roles

4Execute workflow

5Capture evidence

6Report health

98%Test health

7mAvg runtime

4Parallel suites

Automation command lane

Role-aware execution for enterprise QA pipelines.

Boot browser contextdone

Run smoke validationdone

Check permissionsrunning

Capture visual diffsqueued

INTERACTIVE AI SEARCH DEMO

A semantic search preview tied to real project context

The search experience now reflects the actual product surfaces in this repository instead of generic placeholder copy.

enterp

Product workWeaviate UI: browser workbench

A focused interface for schema inspection, object browsing, batch inserts, and GraphQL querying.

similarity 0.99

Tooling surfaceVS Code extension distribution

The same workbench synced into a webview extension with packaging and release automation.

similarity 0.96

Delivery pathsChrome extension + container delivery

Browser extension packaging and a containerized deployment path for local and public use.

similarity 0.94

AI response streaming

Retrieved context aligns with the query intent and surfaces the actual workbench, extension, packaging, and delivery paths in the repo.

ABOUT

Harish Kaparwan

AI engineer focused on vector retrieval systems, agent tooling, and practical delivery paths that turn research-grade ideas into usable products.