I build the AI your business actually needs — not the AI everyone's talking about.

1. 01
   Customer Support & AI Automation

   Agentic AI Customer Support Assistant

   The problem

   Customer support involves more than answering FAQs. Teams need to check orders, review account details, process requests, and follow business policies. When that information lives across separate systems, customers wait longer and often repeat themselves during handoffs.

   The approach

   The system connects an agentic AI assistant to a searchable knowledge base, customer account data, and business APIs. It understands intent, plans the next step, retrieves relevant policies and documentation using RAG, and selects the tools needed to handle each request. Confidence, policy, and risk checks determine whether it can resolve the request. When human support is needed, it passes along a summary and conversation history so the agent can continue with context.

   What it taught me

   Reliable support automation depends on knowing when to answer, when to act, and when to involve a person. Relevant knowledge, account context, and clear escalation rules make that decision possible.

2. 02
   Customer Communication

   WhatsApp Automation

   The problem

   A growing business can’t keep up with WhatsApp inquiries by hand. Replies get slower as volume grows, context from earlier messages gets lost between shifts or staff handoffs, and the same handful of questions get answered from scratch every single time, costing hours that could go toward actually running the business.

   The approach

   Built an automation layer that tracks full conversation context per customer and per thread, so a reply always accounts for what was already said. Messages route through defined workflows (order status, FAQs, escalation to a human) based on intent, tuned to keep replies fast without sounding like a scripted bot.

   What it taught me

   The hard part isn’t generating a reply, it’s knowing what the customer already told you three messages ago. Context management, not response quality, is where most WhatsApp automations actually break, and it’s the part most off-the-shelf chatbot tools get wrong first.

3. 03
   Operations

   Order Delivery Automation

   The problem

   Manually moving orders from "placed" to "delivered" (status updates, handoffs between systems, customer notifications) eats hours every week and is exactly where human error creeps in: a missed update, a forgotten notification, an order that quietly stalls.

   The approach

   Automated the order-to-delivery pipeline end to end. Status changes trigger the next step automatically, customers get notified without anyone typing a message, and exceptions (a stuck shipment, a failed delivery attempt) get flagged for a human instead of silently falling through.

   What it taught me

   Automating the happy path is the easy 80%. The real engineering work is in handling the exceptions (delayed shipments, address changes, cancellations) without the system silently failing or leaving a customer in the dark.

4. 04
   Research & Analysis

   Research Automation

   The problem

   Literature review and gap-finding for research work is slow, repetitive, and easy to do inconsistently. The same search, read, and summarize loop repeats across dozens of sources, and it’s easy to lose track of what’s already been covered versus what’s actually new.

   The approach

   Built a system that structures the research loop itself: pulling and organizing sources, surfacing where the literature actually disagrees or falls short, and producing a structured starting point, not a finished answer, instead of a blank page.

   What it taught me

   AI is good at summarizing a single paper. It’s much harder, and much more valuable, to get it to reliably spot where a field has a genuine gap versus where a topic just looks unexplored because nobody searched for it properly.

5. 05
   Productivity

   Personal Assistant AI

   The problem

   Running client work, freelancing, and a research track at the same time means deadlines and goals live in too many places to track by memory alone. Something eventually slips, usually the thing that felt like it could wait.

   The approach

   Built a personal assistant system that tracks project deadlines and goals across everything in progress, keeps priorities visible in one place, and actively surfaces what needs attention right now instead of listing everything with equal weight.

   What it taught me

   A tracking system only earns trust if it tells you what to ignore, not just what to do. Most task tools fail by showing everything with equal weight, which just moves the overwhelm from your head onto a screen.

These projects are a sample, not the full list. See more of my work →