In progress
RAG Evaluation Framework
A retrieval-augmented QA system with an evaluation harness that actually scores itself — retrieval quality (recall@k, MRR) and answer groundedness — comparing dense against hybrid retrieval with rerankers.
Focus
Mathematical optimization
Formulating real scheduling problems as MILP models and heuristics, then making them fast enough to run in production. C++ on Linux, commercial solvers, profiling.
Applied AI & ML
Retrieval-augmented generation in production on AWS Bedrock, with a bias toward measurable quality — retrieval metrics and groundedness over demos.
Systems that stay up
Monitoring, replication, and failover for software airlines depend on daily. Reliability is a design constraint, not an afterthought.
Experience
-
APR 2023—NOW
Software Developer — Optimization, AI & ML Chapter
Skywise, an Airbus Company
Design and ship crew scheduling optimization for PBS (Preferential Bidding System) in C++ on Linux. Cut solver runtime ~15% with an engine simulator and profiling-guided optimization; instrumented monitoring supporting 99.97% job reliability; built and operate a production RAG assistant on EC2 and Bedrock used by internal teams and airline customers.
-
AUG 2022—DEC 2022
Software Engineer — Full-Stack
Competers Inc.
Built features across a Node.js / Angular / Express stack, migrated legacy SOAP microservices to REST on .NET, and re-architected services as AWS Lambda functions to improve scalability and cut infrastructure cost.
-
MAY 2019—AUG 2019
Software Engineering Intern — Backend
bKash Limited
Built secure Django REST APIs with token-based authentication on one of the world's largest mobile financial platforms, and integrated a Dialogflow agent backed by AWS ElastiCache for low-latency replies.
-
JUN 2018—AUG 2018
Data Analyst
icddr,b
Analysis work at one of the world's leading international health research institutes.
Projects
Selected work from github.com/n5hossai
In progress
Transformer LLM from Scratch
A GPT-style model written from first principles in PyTorch — attention, positional encodings, training loop — then tuned for inference with a KV-cache and quantization, and served as an MCP tool.
Post Feed
A social media feed built on the MEAN stack and deployed to AWS Elastic Beanstalk, covering auth, image uploads, and paginated feeds end to end.
Daily Taskmate
A task manager using Django's MVT architecture with a PostgreSQL backend, deployed on Heroku.
Protein Folding Accuracy
A statistical model scoring how closely a computer-generated protein structure matches a known benchmark — regression and model evaluation on real structural data.
Stratego
A full implementation of the board game in C++, built around object-oriented design patterns and game-state management.
Toolkit
- Languages
- C++ · Python · C# · Perl · SQL · TypeScript · Bash
- Optimization & ML
- MILP & heuristic solvers · OR-Tools · LLM / RAG · AWS Bedrock · PyTorch · FAISS · retrieval evaluation
- Cloud & tooling
- AWS (EC2, Lambda, Bedrock) · GitLab CI/CD · Docker · JFrog · Nagios · MariaDB · MongoDB · Linux
Education
-
2017—2021
Bachelor of Mathematics, Honours — Computational Mathematics
Minors in Statistics, Computer Science, and Combinatorics & Optimization, with a depth in Physics.
Certifications AWS Certified Developer – Associate Supervised Machine Learning
Open to work in optimization, operations research,
and applied AI/ML engineering.
Waterloo and Toronto, or remote across Canada.