Tanzim Hossain Romel

Me in front of Lake Minnewanka, Banff National Park, September 2026.

I study how AI agents behave in real software systems, and how to make their decisions inspectable, reliable, and trustworthy.

I bring about three years of professional software-engineering experience, formerly at IQVIA.

My research interests are AI4SE, LLM4Coding, trustworthy AI, long-horizon coding agents, and AI for SRE.

I care about focused, sustained work. Here is how I think about work ethic.

I live in Edmonton, Alberta. Fun fact: Edmonton’s North Saskatchewan River Valley is one of the largest stretches of connected urban parkland in North America.

Research

I work on three connected questions.

Long-horizon coding agents. How can they preserve context, repository memory, and architectural intent as software evolves? See ContextLedger.
AI for SRE. How should agents be evaluated when diagnosis and recovery happen inside live reliability conditions? See SREGym.
Trustworthy AI. How can we inspect agent choices, security boundaries, and evidence when the final answer still looks acceptable? See SHIFT.

Education

University of Alberta, M.Sc. in Computing Science, September 2026 to present. U-A-Goose group, supervised by Dr. Zhou Yang.
Bangladesh University of Engineering and Technology (BUET), B.Sc. in Computer Science and Engineering, April 2018 to May 2023.

Selected projects

ctxhelm

Local-first context plans, release proof, and source-free agent evaluation.

ContextLedger

A research notebook on OpenCode compaction, provenance, and live GPT-5.5 experiments.

PatchSmith

System design and R&D notes for bounded patches, sandbox validation, and auditable repair runs.

1BRC C# on Apple Silicon

A .NET 10 NativeAOT performance case study: byte parsing, native aggregation, and ARM64 tuning.

All projects

news

Sep 27, 2026	SREGym blog: how we turned real-world postmortems into reproducible SRE benchmark problems. Blog · Tianyin’s post
Sep 01, 2026	I have started graduate study in Computing Science at the University of Alberta, in the U-A-Goose group. Link
Jun 30, 2026	‘An Empirical Study on Remote Code Execution in Machine Learning Model Hosting Ecosystems’ is under review at the International Conference on Software Engineering (ICSE) 2027. Preprint available on arXiv. Link
Jun 30, 2026	Submitted ‘The Choice Can Be the Attack: Auditing Aligned Backdoors in LLM Agents’ to the Transactions of the Association for Computational Linguistics (TACL) 2026 with Chowdhury Rakin Haider. Link
Jun 01, 2026	Started a remote summer research internship at the University of Illinois Urbana-Champaign through UIUC++ Summer Research in Software Engineering (SRSE) 2026 with Professor Tianyin Xu. Link

latest posts

Sep 29, 2026	Working Hard Means Working Deep and Resting on Purpose
Jun 18, 2026	Building a fast 1BRC solver in C# on Apple Silicon
Jun 01, 2025	Memory Management in Production: Avoiding the Silent Killers

selected publications

ICSE 2027
An Empirical Study on Remote Code Execution in Machine Learning Model Hosting Ecosystems

Mohammed Latif Siddiq, Tanzim Hossain Romel, Natalie Sekerak, and 2 more authors

2026

Under review at the International Conference on Software Engineering (ICSE 2027)

Abs arXiv Bib PDF

Large-scale cross-platform security study of roughly 45,000 repositories across five ML model hosting ecosystems, measuring unsafe deserialization, eval injection, malware signals, and recurring developer security misconceptions.
@misc{siddiq2027rce, title = {An Empirical Study on Remote Code Execution in Machine Learning Model Hosting Ecosystems}, author = {Siddiq, Mohammed Latif and Romel, Tanzim Hossain and Sekerak, Natalie and Casey, Beatrice and Santos, Joanna C. S.}, year = {2026}, note = {Under review at the International Conference on Software Engineering (ICSE 2027)}, }
TACL
The Choice Can Be the Attack: Auditing Aligned Backdoors in LLM Agents

Tanzim Hossain Romel and Chowdhury Rakin Haider

2026

Submitted to TACL 2026; manuscript not public

Abs Bib

LLM agents can be backdoored without visibly failing the task: a trigger can steer an agent toward an attacker’s preferred valid option while the final answer still looks acceptable. SHIFT (Structured Hidden Influence Test) reruns matched tasks with and without a known trigger and checks whether the changed choice favors the attacker’s target after accounting for ordinary option quality such as price or rating.
@misc{romel2026shift, title = {The Choice Can Be the Attack: Auditing Aligned Backdoors in {LLM} Agents}, author = {Romel, Tanzim Hossain and Haider, Chowdhury Rakin}, year = {2026}, note = {Submitted to TACL 2026; manuscript not public}, }
CSCW
Sentiment Analysis of Anonymous Crisis Reports in Bangladesh

Tanzim Hossain Romel

2025

Submitted to the ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW)

Abs Bib PDF

Transformer-based analysis of 124 anonymous crisis reports from Bangladesh’s 2024 national crisis, using a multilingual pipeline for Bengali and Romanized Bengali to study crowd-sourced sentiment in a high-risk setting.
@misc{romel2025ureporter, title = {Sentiment Analysis of Anonymous Crisis Reports in Bangladesh}, author = {Romel, Tanzim Hossain}, year = {2025}, note = {Submitted to the ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW)}, }