Applied AI · Agentic Systems · Multimodal AI · Enterprise Software
I build AI systems where the problem is not neatly defined yet.
My work sits between customers, product, and engineering: understanding messy business workflows, identifying where AI can create leverage, designing the architecture, building the system, integrating it into existing operations, and getting it running in production.
I have spent much of my career building software directly around real business constraints, from enterprise operations and financial workflows to AI agents, semantic memory, computer vision, retail experiences, and sales automation.
Today, I am building iKawn and working primarily on agentic systems, multimodal AI, MCP, AI-native enterprise workflows, and forward deployed engineering.
Persistent intelligence infrastructure for AI agents
An internal AI operating layer that gives agents persistent memory, tools, documents, background workers, semantic search, and controlled access to business operations.
Architecture
- Node.js / Express
- PostgreSQL + pgvector
- Claude agent runtime
- Gemini embeddings
- MCP-compatible tools
- Background workers and schedulers
- Human approval workflows
- Multi-provider AI routing
- Fly.io
OpenBrain explores one of the problems I care most about:
How do you turn stateless AI models into reliable systems that accumulate context, take action, and operate safely over time?
AI virtual try-on for physical retail
A multimodal retail platform that lets shoppers visualize themselves wearing different products through interactive kiosks and browser experiences.
I worked across the entire system, including:
- Virtual try-on product architecture
- Retail and kiosk UX
- Buyer-facing demo experience
- Lead qualification and routing
- Pricing and commercial packaging
- Web infrastructure
- Retail deployment workflows
Mirror is an example of how I approach Forward Deployed Engineering: the technical problem cannot be separated from store behaviour, hardware constraints, customer experience, latency, economics, and how retailers actually buy technology.
Live: ikawn.com/mirror
Production business administration system
A live internal operating system built around the actual workflows of Rameshwar Enterprises.
The platform handles:
- Payroll
- Employee attendance
- Invoicing
- Purchase invoices
- Purchase orders
- GST / TDS workflows
- Travel tickets
- Banking exports
- Mobile attendance APIs
- Operational reminders
I also integrated Gemini-based document intelligence for extracting ticket information from PDFs and matching it against employee data.
This system represents a different side of FDE work: extending and modernizing a live business-critical system without disrupting the workflows people already depend on.
Making traditional software agent-accessible
A ProcessWire module that exposes CMS operations through an agent-oriented JSON-RPC interface.
AI agents can discover, read, create, update, and manage content while respecting the application's existing authentication and permission model.
Repo: github.com/vineonardo/ProcessMCP
AI-native sales workflow automation
A SaaS system that turns a short business questionnaire into:
- Ideal Customer Profiles
- Target lead lists
- Email sequences
- Outreach content
- Social content ideas
Built using Next.js, TypeScript, Supabase, Claude, Apollo, and Stripe.
The interesting problem here was not content generation itself, but converting loosely structured business information into structured, usable sales actions.
I usually enter projects before a clean specification exists.
The workflow is closer to:
Understand the business → identify leverage → prototype → deploy → observe → iterate
I am comfortable moving between:
- Customer conversations
- Product strategy
- System architecture
- AI orchestration
- APIs and integrations
- Databases
- Full-stack engineering
- UX
- Infrastructure
- Deployment
- Commercial constraints
I care less about whether something is called an "AI product" and more about whether it meaningfully improves the underlying workflow.
- AI agents and agent orchestration
- Persistent AI memory
- MCP and tool ecosystems
- Multimodal and real-time AI
- Computer vision
- Human-in-the-loop systems
- AI-native enterprise software
- Local and edge AI
- Agentic commerce
- Reliable production AI systems
A large part of my current work lives in private repositories because it involves customer systems, internal infrastructure, or commercial products.
The public repositories here are selected examples rather than a complete representation of what I build.
iKawn: ikawn.com