Docket
Receipt tax tracker
Photograph a receipt, get ATO-categorised deductions, offline.
Selected work
Things I built
Client code is private. These are mine.
AI Vision App
Edge object detection
Real-time object detection on-device with YOLOv8 and TFLite.
Agent Pipeline
Agent workflow template
Seven-subagent template: spec, implement, review, test.
Monte Carlo Simulator
News-adjusted simulation
Portfolio simulation adjusted by live news sentiment.
Also built
Shader practice
Jellyfish Aquarium
Language interpreter
Scala Interpreter
Note-taking platform
DevNotes
How the work arrives
Three I was sent into. Three nobody asked for.
Embedded in six client teams.
- Client engagements
- 6
- Tickets delivered
- 130+
- Records migrated
- 1.5M
- Less data entry
- 80%
Sent in
Three of six engagements. Clients unnamed.
Mobility & vehicle services
1.5M records migrated
Python · FastAPI · Next.js · SQL
Replatformed onto a custom site I built, records included, zero service loss.
National health insurer
Manual entry down 80%+
Python · FastAPI · Computer vision · AWS
Automated claim capture for NDIS providers.
Financial services
130+ tickets resolved
Apex · LWC · SOQL · Workato
Platform built across multiple releases.
Built unasked
Nobody assigned these. I found the gap, built it, got it adopted.
PromptBench
30+ users · 2hrs → <30min
TypeScript · Node.js · MCP · Claude API
Now the default prompt check across delivery teams.
Expertise Finder
Python · FastAPI · RAG · Claude API
Searchable directory of who delivered what.
QA generation
AWS Lambda · Bitbucket Pipelines · LLM
Test cases exported to the source ticket.
Case studies
Three problems, and how I solved them.
Names withheld. Numbers are real.
01Mobility & vehicle services
Replatform a live business without stopping it
- Situation
- Move a client onto a custom website I built, records included, while the business kept trading.
- Constraint
- Live business, embedded with the client team. No service loss allowed.
- Decision
- Built the site and migrated every record alongside the client team.
Python · FastAPI · Next.js · SQL
Result
1.5M
Records migrated
0
Service loss
Full
Platform build
02National health insurer
Take the typing out of NDIS claims
- Situation
- NDIS providers keyed claims by hand.
- Constraint
- Messy formats. Data feeds real payments.
- Decision
- OCR with async workers, so capture runs in the background.
Python · FastAPI · Computer vision · AWS
Result
80%+
Less manual entry
Async
Capture pipeline
03Internal tool, built unasked
Make prompt changes provable, not a gut feel
- Situation
- Teams tuned prompts by eye, with no before/after check.
- Constraint
- Unassigned. Had to be easier than skipping it.
- Decision
- Before/after evaluation tool on the Claude API, with MCP support.
TypeScript · Node.js · MCP · Claude API
Result
30+
Engineers using it
2h → 30m
Iteration time
Play with it
I wrote a language once. Here it is.
Ported from Scala. Type into it.
Lisp REPL
A small Lisp, running in your browser.
Tokeniser, parser and evaluator, no libraries.
Try an example below, or type something.
›
Play with it
A shader that got away from me.
One fragment shader. Drag through it, click to send a shockwave.
Raw WebGL2. Trails come from a feedback buffer.
Work log
Release notes, not a résumé.
v2.0 · live02/2024 - PresentSydney, AU
Associate Software Developer
Ingenuity Partners
- Delivered technical designs across 6 client engagements by embedding with engineering, product and business teams and leading pre-build design discussions.
- Eliminated stored credentials between Workato and Salesforce by implementing Microsoft Entra ID app-to-app auth with the OAuth 2.0 client credentials flow.
- Cut manual data entry by over 80% for a national health insurer by building an OCR pipeline that extracts claim data from provider receipts.
- Migrated 1.5M records off a legacy CRM with zero service loss by building a replacement full-stack system in Next.js, TypeScript and Python.
- Connected client platforms to external systems by building REST, AWS and Workato integrations with authentication, error handling and data reconciliation.
- Resolved 130+ tickets across configuration, integration and custom development to deliver a full platform implementation for a financial services client.
- Cut prompt iteration time from 2+ hours to under 30 minutes for 30+ engineers, BAs and delivery leads by designing and shipping an internal LLM benchmarking tool.
- Shipped a multi-agent compliance analysis service in Next.js, Python (FastAPI) and LangGraph.
v1.0 · released01/2023 - 02/2024Sydney, AU
Software Engineering Intern
IBM
- Led the chaincode development team on the Open Healthcare Platform by IBM under the supervision of mentors from IBM and Macquarie University.
- Gained experience in Java OOP, functional programming features, Junit, Gradle, Ansible, Minifabric and Hyperledger Fabric.
- Used virtual environments and Docker containerisation to develop the smart contracts and run the network.
- Built the codebase for the blockchain infrastructure, leveraging Hyperledger Fabric and Minifabric.
- Developed code to the HL7-FHIR Standard for the healthcare industry.
- Handled DevOps for software development using Ansible Playbook.
- Implemented Agile methodology and GitHub VCS for project management.
- Wrote SDLC documentation for the tasks and assigned tasks to team members for seamless progress of the project.
- Tested and ran QA on the developed code using JUnit tests and other whitebox testing.
EducationJune 2024
Bachelor of Information Technology
Macquarie University
Double major: Software Technology and Data Science
What I build
Three stacks, one discipline.
Web & mobile
React, Next.js and React Native over Python and Node, on AWS.
LLM-integrated systems
Retrieval, agent workflows, evaluation harnesses, document extraction.
Enterprise platforms and solutions
Salesforce and Workato: event-driven Apex services and Lightning Web Components for live client workflows.
Stack
Tools
Bold is what I use daily.
Languages
- TypeScript
- JavaScript
- Python
- SQL
- Java
- Apex
- Dart
Frontend
- React
- Next.js
- React Native / Expo
- Tailwind / Sass
- Redux / Zustand / React Query
- Jest / React Testing Library
- Three.js
- Flutter
- Lightning Web Components
- Framer Motion
Backend
- FastAPI
- Node.js
- Express / NestJS
- .NET
- REST / GraphQL / gRPC / SOAP
- WebSockets
- Kafka & event-driven architecture
AI & LLM engineering
- Claude Code
- RAG pipelines
- Anthropic Claude API
- Model Context Protocol
- LangGraph
- Multi-agent orchestration
- Prompt evaluation
- Spec-driven development
- YOLOv8 / TensorFlow Lite
Platforms & tooling
- AWS, Lambda, API Gateway, S3, Secrets Manager
- Git / GitHub
- GCP
- GitHub Actions
- Docker
- Serverless Framework
- Salesforce (Apex, LWC, Flows)
- Workato
- Firebase
- Vercel
- Microsoft Entra ID (Azure AD)
- OAuth 2.0
- Google Cloud Platform
- Microsoft Power Platform
Data & practices
- SQLite
- SOQL
- PostgreSQL
- MongoDB
- DynamoDB
- DuckDB
- Agile delivery
- Solution design
- Automated testing
- CI/CD
The constraint run
Six snippets. One flaw each.
Click the flawed line.
01 / 06
ApexSalesforce runs every transaction against a hard query ceiling.
Click the line you'd flag in review.
And this is what it costs.
Salesforce allows 100 queries per transaction. Go over and it rolls back.
1100200300400
Query in loop
for (r : records) { SELECT … }
SOQL queries50 / 100
CPU time1.2s / 10s
Within budget. Holds until 100 records.
Bulkified
SELECT … WHERE Id IN :ids
SOQL queries1 / 100
CPU time0.1s / 10s
Flat. One query at 50 records, same as at 1.
Note: real Apex limits, illustrative timings.
Hiring? Let's talk.
Open to full-stack, applied AI and Salesforce roles in Sydney. Hybrid, remote or in-person. Happy to walk through any of this.
rakeenabtahee@gmail.com