Work

Open SourceGitHub, npm, PyPI, and public packages.

Depth

JourneyCareer timeline, roles, and education.AchievementsScholarships, rankings, and milestones.

Writing

PublicationsResearch-facing work and technical reports.ProfilesAll key public links in one place.

Selected Work · 01

The work,with the lid off.

From enterprise observability and AI assistants to marketplaces, blockchain systems, and finance-oriented prototypes — each one shipped under a real constraint, not staged for a portfolio.

Layer

What people touch

Open the case studiesHow I build

Before any of these shipped

Forty things off a brief, and the five it turned out to be about.

No brief arrives as a specification. It arrives as 40 fragments — asks, constraints, and the thing mentioned in passing that ends up deciding the architecture — and the work is finding out which pane each one belongs to. 28 of them seat. 12 are trade-offs, and a trade-off is a call somebody has to make, so they stay out with the question attached.

State40 shards loose

  1. 01MONEYledgers, payouts, reconciliation

    payouts have to reconcile to the cent · refunds arrive out of order · three currencies, one ledger · the provider retries silently

  2. 02SCALEtraffic that kept growing

    it was fine at ten thousand · the queue backs up at 9am · one endpoint is 80% of traffic · the database is the bottleneck

  3. 03LEGACYcode nobody wanted to touch

    nobody has run the tests in a year · two systems both call themselves the source · the last engineer left in March · it is business-critical and undocumented

  4. 04SIGNALdata somebody has to act on

    the dashboard is decorative · two teams report different totals · we need it before the board meeting · product and finance disagree on the number

  5. 05HANDOVERwhat stays after I leave

    two people have to maintain it · in six weeks · tests that still run next year · the runbook has to fit on a page

  • Open 01exactly once, or sub-second — which one at 9am?“exactly once, always” vs “sub-second, under load” — belongs to two panes
  • Open 02cache the hot path, or trust the ledger?“sub-second, under load” vs “exactly once, always” — belongs to two panes
  • Open 03which of the two is the source of truth?“two systems both call themselves the source” vs “product and finance disagree on the number” — belongs to two panes
  • Open 04rewrite it, or keep it alive another year?“rewrite it” vs “do not rewrite it”
  • Open 05six weeks, or done properly?“in six weeks” vs “done properly”
  • Open 06who is on call for this once I am gone?“who is on call after handover” vs “the team is two juniors and a contractor”
Loose
40 shards, tumbling, catching light at random depths.
Seating
Each finds its socket. The joins between them stay visible.
Five panes
Whole, and etched with what each one turned out to be about.
Six open
12 will not go in. The open sockets are the deliverable.
  • AI Systems
  • Cloud Architecture
  • FinTech
  • Web3
  • Data Engineering
  • Marketplaces

Project library

Browse all projects

A deeper look into the architecture, challenges, and results of selected builds.

Showing 8 of 8 projects

AI

Hermes Nexus AI-Native Deal Execution Graph

A graph-shaped AI-native workspace connecting procurement, contracting, and compliance evidence so commercial conversations can be traced into execution-ready workflows.

  • JavaScript
  • TypeScript
  • React
  • Next.js
  • Python
  • +4

View case study

AI

HeimdallAI Autonomous Penetration Testing Platform

An AI-agent-driven security assessment platform built under hackathon constraints, focused on vulnerability detection, monitoring, and explaining findings instead of just flagging them.

  • JavaScript
  • TypeScript
  • React
  • Next.js
  • Gemini
  • +3

View case study

Blockchain

Web3 Smart Asset Platform

A final-year project exploring smart tokens that enable buying, selling, lending, and time-bound trading of gaming assets with profit-sharing mechanics.

  • Solidity
  • TypeScript
  • React
  • Next.js
  • HardHat
  • +3

View case study

Cloud

Real-Time Notification Migration

Migrated notification workflows to SQS, Lambda, and Firebase while improving backend reliability and pushing a large JavaScript codebase toward TypeScript and server-compatible React patterns.

  • SQS
  • AWS Lambda
  • Firebase
  • TypeScript
  • React
  • +1

View case study

Cloud

Legacy Monorepo and Microservice Modernization

Modernized a legacy codebase into a clearer monorepo and microservice setup using Yarn workspaces, Turborepo, Docker, AWS SAM, and stronger development standards.

  • Yarn Workspaces
  • Turborepo
  • Docker
  • AWS SAM
  • Lambda
  • +1

View case study

AI

Enterprise AI Assistants with Guardrails

Assistants that query backend systems, databases, and user-defined workflows, kept inside explicit operational constraints instead of improvising.

  • OpenAI
  • Vector Store
  • TypeScript
  • Python
  • Node.js
  • +1

View case study

Mobile

CarHub B2C and B2B Marketplace

Responsive web, mobile, and backend systems for a vehicle marketplace serving consumer and business traffic across products, services, and wholesalers.

  • React
  • React Native
  • Next.js
  • Node.js
  • MongoDB
  • +2

View case study

Data

AppNavi Observability Platform

A distributed observability and event analytics platform for enterprise clients, processing large event streams into dashboards operators can actually act on.

  • C#
  • .NET Core
  • AWS Lambda
  • Kinesis
  • S3
  • +4

View case study

Working Stack

Technologies behind the builds

A cross-section of tools used to deliver production systems.

Software Engineering

3

Languages, frameworks, and the craft of building and shipping full-stack products.

  • Python
  • React
  • TypeScript

Cloud, Data & DevOps

2

Infrastructure, pipelines, and the operational systems that keep distributed software reliable.

  • PostgreSQL
  • Solidity

Other Topics

2

Additional themes that come up across the work.

  • AWS
  • Next.js

Methodology

Rigorous execution for complex requirements

Durability and performance matter, but only if the system still serves the business goal it was built for.

Contact for Collaboration

  1. First-principles architecture

  2. Iterative, feedback-driven delivery

  3. Emphasis on observability and scaling

Lifecycle

How projects move from concept to code

A structured path from discovery to deployment.

Step 01

Discovery

Get specific about what success actually looks like before writing any code.

Step 02

Architecture

Design around the load and failure modes the system will actually hit.

Step 03

Implementation

Build the riskiest part first, then everything else gets easier.

Step 04

Deployment

Ship it, watch how it behaves under real traffic, and fix what only shows up there.

Next Steps

Continue exploring

Why Most AI Agent Pilots Never Reach Production (And How to Be the Exception)

MIT found 95% of AI pilots fail to move the P&L, and Gartner expects 40% of agentic AI projects canceled by 2027. Here's what the surviving few do differently.

Navigate

TypeScript Patterns That Prevent Real Production Bugs

Master TypeScript patterns that catch bugs at compile time. Discriminated unions, branded types, and exhaustiveness checks prevent the production failures that slip past runtime testing.

Navigate

Vector Databases Compared for Production RAG Systems

Evaluate Pinecone, Weaviate, Qdrant, pgvector, and Milvus for your production RAG pipeline. Technical comparison of algorithms, scaling strategies, and cost tradeoffs.

Navigate

How to Prepare for Full-Stack and Agentic AI Engineering Interviews

Master the interview process for full-stack AI roles. Learn what hiring managers actually test, from agent architecture to production failure modes.

Navigate

Where to go next

More case studies

The rest of the record