- Plate I
- Profile
- Degree
- MEng CS & SE, Birmingham
- Role
- Financial Analyst, Amazon Leo
- Location
- United Kingdom
- Updated
Financial Analyst in Amazon's satellite broadband business.I build software there, and on my own time.
FIG. 1Amazon Leo
I am a Financial Analyst on a twelve month industrial placement at Amazon, inside Amazon Leo (formerly Project Kuiper), Amazon's low earth orbit satellite broadband business, building a constellation of more than three thousand satellites.
The finance work is genuinely quantitative, and the engineering work is real. Both are listed below, in the order I actually spend time on them.
- 01
Took ownership of an AWS hosted internal deal approval platform used by 100+ employees across four functions. Introduced source control and a documented production baseline.
- 02
Rebuilt its deal intake and approval flow to production standard, aligning legal, business, tax and finance on a single view of every deal.
- 03
Build internal tools supporting finance workflows and financial modelling.
- 04
Write SQL and Python to pull and transform data, and present the findings to stakeholders. Produce P&Ls and forecasts for deals.
- 05
Run internal sessions helping other employees use AI tooling more effectively.
Note. Figures beyond the 100+ platform users are internal and are not published here. The Leo name and the Kuiper rebrand are both public.
1 The business the role sits inside. I work in its finance org; I do not build spacecraft.
2 The deal approval platform, drawn as what it does.
FIG. 2Trip Cost & Time Decomposer
Breaks down the real cost and time of a trip, with explicit assumptions and a confidence level on every estimate.
The estimation engine is the point. A geodesic distance and regional multiplier model produces a usable number with no upstream provider at all, and every estimate carries a confidence level saying which path produced it. A live flight provider can sit in front of it as the primary source, with Resilience4j circuit breaking the swap. The Amadeus integration it shipped with has stopped serving, so it now runs entirely on the fallback path, which is the case it was built for. The whole stack, API, frontend, worker and database, runs from a clean clone under Docker Compose.
- 1 Next.js 14 client, TypeScript
- 2 Spring Boot 3 API, Java 21
- 3 Python ETL worker seeding city and cost of living data
- 4 Resilience4j circuit breaker and confidence tagging
- 5 PostgreSQL 15
FIG. 3Banking Multi-Agent System
A customer relationship assistant for a retail bank where a root agent delegates to specialised sub-agents rather than doing everything in one prompt. Routing is a model tool call, not hand written branching.
Responses are grounded in BigQuery so the agent answers from transaction records instead of improvising. Each agent's system prompt lives in its own file, loaded at construction, and an ADK evalset checks prompt changes against expected tool call behaviour.
- 1 Root agent, routing by model tool call
- 2 Complaints sub-agent, exposed as a tool
- 3 Navigator sub-agent, exposed as a tool
- 4 BigQuery, grounding every answer in records
Scope. This is a build on Google's Agent Development Kit using their fictional Cymbal Bank sample data. It is not an original agent framework and there is no real customer data anywhere in it.
FIG. 4Robotics SLAM & Simulation
A modular localisation and motion control pipeline for an e-puck robot in simulation. Particle filter localisation estimates position from noisy sensor data, with odometry calibration and waypoint navigation on top.
The part I would point at is the tooling around it: automated logging and replay so a run can be analysed after the fact, and error propagation measured rather than guessed at.
- 1 Webots simulation environment
- 2 Sensor processing
- 3 Particle filter localisation
- 4 Waypoint navigation and odometry calibration
- 5 Automated log replay and test harness
Coursework, not published
FIG. 5Tools I run daily
Not portfolio pieces. These are tools I built because I wanted them, and they run every day on my own machine.
Job Scout
Reads my inbox each morning, scores roles against a preferences file, and builds a tailored LaTeX CV for anything scoring above the threshold. Python, SQLite, launchd.
LocalFlow
Push to talk dictation running entirely offline. Parakeet MLX for transcription, a local model for cleanup.
London Scout
A flat hunting map that scores areas by real TfL commute time rather than distance, drawn as hexes over listings data.
Local Hub
A dashboard on localhost that starts at login and launches the rest of them.
PLATE IIRecord
- Degree
- MEng Computer Science & Software Engineering, University of Birmingham. September 2023 to June 2028, including the twelve month industrial placement.
- Award
- First place, Siemens Birmingham Project, June 2024. Led a team of seven against more than ten competing teams, and delivered the final presentation to 200+ attendees.
- Society
- Treasurer, University of Birmingham Chess Society, June 2024 to present. Budgeting, reimbursements and reporting for 60+ members. Member of AFNOM, the cybersecurity society.
- Languages
- Arabic, native. French, fluent. English, fluent.
PLATE IIIGet in touch
I am looking at summer 2027 internships and 2028 graduate roles, in software engineering, backend, cloud and infrastructure, machine learning, and technology inside finance.
- Emailmekhzerramy@gmail.com
- GitHubgithub.com/Ramy3077
- LinkedInlinkedin.com/in/ramymekhzer