AI Research Engineer · Toronto

I build reliable AI agents that earn their place in production.

I'm an AI Research Engineer working on agent evaluation, grounded tool use, retrieval, memory, failure recovery, and auditable AI workflows.

View agent case studies Download résumé

GitHub LinkedIn Scholar

Agent reliability stackmeasured

ContextGrounded retrieval

ActionTools + orchestration

ControlConstraints + recovery

EvidenceEvaluation + traces

Ph.D.Computer Science

3+ yearsApplied AI research

AAAI · Canadian AIPeer-reviewed venues

3 sectorsIndustry · government · academia

Selected case studies

Built to answer a hard question — and to show the answer.

All case studies

Preprint + public research artifact2026
Project Ariadne
An agent's written reasoning is not evidence that the reasoning caused the answer.
My role: Designed the intervention protocol, built the auditing harness, and ran the evaluation.
- Counterfactual replay
- Trajectory logging
- LLM evaluation
- Python
23 of 30
audited trajectories showed a faithfulness violation under the counterfactual-intervention protocol
This is an audit finding about the agents under test, not a failure rate of the tool. 23/30 is 76.7%.
Read the Project Ariadne case study Paper for Project Ariadne Code for Project Ariadne
Prototype architecture and case study2025
AgentMesh
A single LLM asked to build software stops at the first thing it gets wrong.
My role: Sole author. Designed and implemented the framework.
- Multi-agent systems
- Tool use
- Self-repair
- Python
Self-repair
execution feedback drives automatic repair — demonstrated end to end, not benchmarked
No benchmark, baseline, or success rate is reported in the paper, so no comparative claim is made here.
Read the AgentMesh case study Paper for AgentMesh
Peer-reviewed research2023–2026
Search & planning
When a heuristic gives a search no gradient to follow, the search stalls in a region it cannot see its way out of.
My role: Co-developed the restarting-random-walk planner (AAAI 2026). Lead author on the separate refactoring-heuristics work.
- Heuristic search
- AI planning
- PDDL
- Admissible heuristics
70.3%
of unbounded-region tasks, where the restarting-random-walk planner was faster than or uniquely successful against breadth-first search
The 70.3% is measured on unbounded-region tasks specifically, not on planning benchmarks as a whole. "Uniquely successful" means the comparator did not solve the task at all.
Read the Search & planning case study Paper for Search & planning Paper for Search & planning

Approach

Four things an agent needs before reliability is a claim you can check.

This is how I reason about dependable agent systems — not a universal standard, but the structure I build and evaluate against.

01Context: Retrieve the state the task actually depends on, so an answer is grounded in real data rather than model priors.
02Action: Give the agent tools and an orchestration structure, so it can change the world instead of only describing it.
03Control: Constrain what the agent may do and define what happens when a tool call or a step fails.
04Evidence: Log trajectories and score them, so a claim about reliability can be checked rather than asserted.

Current work

One engineering role, two research appointments.

Full background

May 2026 — present
AI Engineer, Agentic Systems
Flybits · Full-time
Agentic systems that plan, call tools across internal and partner APIs, and complete multi-step tasks with every decision logged for end-to-end auditability.
May 2026 — present
AI Research Engineer, sAIpien
MIT Media Lab · Part-time research affiliation
Perspective-aware agents, privacy-preserving long-term memory, and benchmarks for transparency and human oversight.
May 2026 — present
Postdoctoral Fellow / AI Research Engineer
Toronto Metropolitan University · Part-time academic appointment
Reproducible evaluation harnesses for LLM-agent reasoning, orchestration, context management, and reliability.

Selected research

Peer-reviewed work and preprints, labelled as such.

Full publication record

PreprintarXiv · 2026
Project Ariadne: A Structural Causal Framework for Auditing Faithfulness in LLM Agents
A protocol for testing whether an LLM agent's stated reasoning actually caused its answer: intervene on the reasoning trace, replay the workflow from the modified state, and check whether the output moves.
Peer reviewedAAAI Conference on Artificial Intelligence · 2026
Breadth-First Search vs. Restarting Random Walks for Escaping Uninformed Heuristic Regions
When a heuristic gives a planner no gradient to follow, the planner stalls. This paper compares breadth-first search against restarting random walks for escaping those regions, and shows the random-walk approach needs only linear worst-case memory.
PreprintarXiv · 2025
AgentMesh: A Cooperative Multi-Agent Generative AI Framework for Software Development Automation
A framework in which Planner, Coder, Debugger, and Reviewer agents pass work between each other, using execution feedback to repair their own output rather than stopping at the first failure.

Field notes

Research and engineering notes on reliable agent systems.

All writing

Jun 8, 2026AI · AI Research
A Practical Roadmap for AI Research
A structured guide to the mathematics, machine learning fundamentals, and research practice behind modern AI work.
Jun 8, 2026Linear Algebra · AI Research
Linear Algebra for AI Researchers: Foundations
A focused guide to the vectors, matrices, transformations, and decompositions behind modern machine learning.

Start with the hard part

Building an agent that must work outside the demo?

I work on evaluation, grounded tool use, failure recovery, and auditable agent workflows. Open to research collaborations and selective advisory work in agent evaluation and reliability.

Discuss an agent problem sourena.khanzadeh@gmail.com

Other work

I'm also the founder of Ariadne Growth Systems, a separate growth-engineering company. Not to be confused with Project Ariadne, the faithfulness research above.