Experience
9 roles
Research engineering, fintech, food delivery, and developer tools — in founding-team and IC contexts both. Open a row for the detail; for project-level write-ups see Work.
Roles
Newest first
Agentic LLM workflows for vulnerability detection, exploitation, and patching — co-supervised by Marc Juarez (Lecturer, Security, Privacy and Trust) and Adriana Sejfia (Institute for Computing Systems Architecture), with Dr. Mohammad Tayebi at SFU. Designed and ran a controlled ablation on whether exploit evidence improves LLM-generated security patches, holding the agent, model, and prompt wording constant so the evidence was the only variable: the result was a null one — no significant difference between the exploit-informed and alert-only arms (+3.0 points on a 0–100 scale across 12 matched CVE pairs, win/loss/tie 6–5–1, sign test p ≈ 1.0) at roughly 3× the cost. The contribution is the mechanism diagnosis behind it: a single proof-of-vulnerability is a narrowing signal, so an evaluation gate defined on that one path rewards patches that close it rather than the bug class behind it. Also built the evaluation harness (reference-anchored LLM-as-judge scoring against the real upstream fix across 8 dimensions, K=3 ensemble with medoid selection, aggregates computed in code, hard gates on vulnerability-elimination and regressions, and a measured noise floor so any reported difference can be read against it), curated a linked-CVE dataset and directed graph of security fixes that were themselves later found incomplete, and shipped the AutoSec Lab dashboard (FastAPI + React) for run orchestration and side-by-side arm comparison.
Building the agent layer of an LLM-powered agentic IDE. Shipped: a six-tool subagent system (DelegateTask / SendMessage / WaitForTask / CheckTaskOutput / StopTask / ListTasks), a TDD-style RED→GREEN multi-agent DAG orchestrator with Mission Control dashboard, the prepare→permission→execute tool lifecycle, chain-of-density + KV-cache-aware context management, SWE-Bench and Aider-Polyglot evaluation infrastructure, and an MCP server exposing the test-and-break loop to AI agents. Authored the empirical study behind Fabric’s externally-published March-2026 multi-model orchestration benchmark report, and ran a SWE-bench-with-vs-without-GraphRAG experiment whose negative result informed the team’s no-ship recommendation. Also designed and shipped BDC Market Watch end-to-end — a daily-cycle multi-source news → LLM-judged signals pipeline for the Business Development Bank of Canada (Farpoint client engagement, ~3 weeks idea-to-production), with a local-embeddings-and-clustering layer that cuts Copilot Studio LLM call volume by ~5–10× vs. a per-article baseline.
M.Sc. Computing Science under Dr. Mohammad Tayebi. First-author of ZeroFalse (preprint, arXiv:2510.02534) and ThreatEZ (CMPT 785 course paper, ACM-formatted). Designer of AutoSec (fully agentic vulnerability remediation) and the Golden Repository (LangGraph pipeline producing 89 verified end-to-end CVE reproductions across Python and Java with vulnerable + patched container pairs). TA across Database Systems I, Software Development Methods, Digital Systems Design, Digital Logic. CGPA 4.00/4.33; SFU Graduate Fellowship (Summer 2025).
Evaluated emerging AI/ML capabilities (recommendation, personalization, vision-based form analysis, LLM coaching) for an Iranian fitness / health-tech platform; ran the user-research → spec → engineering-handoff loop and bridged product needs into concrete technical scope.
Co-founded Iran’s leading crypto social-trading and copy-trading platform; funded by Finnova (largest FinTech accelerator in Iran) after a 2-month MVP. Architected the trading engine: smart order routing across 5+ external exchanges (Binance, KuCoin, regional venues), best-execution price aggregation, async Python + Celery, per-exchange adapter pattern over a normalized schema, sub-second price-refresh fan-out via Redis Pub/Sub, and the Copycat replication state machine with slippage controls and per-follower position sizing tracking low-thousands of leader-follower pairs at peak. ~40k users in 18 months; code review and mentorship across a 3–5 engineer team.
Customer Experience team on Iran’s largest food-delivery platform (10M+ users). Built the company-wide Octopus BI layer; adapted Uber’s DeepETA to motorbike delivery (~27% accuracy lift, 24% fewer delays); shipped a 3M-user churn-prediction pipeline (13% monthly churn reduction); rules-based vendor-fraud detection (10% CSAT lift, NPS 5→7); shipped FoodFeed via outsourced engineering.
Co-founded the engineering team of an Iranian freelancing platform; delivered a working beta in 6 months from zero with a 3-person founding team. Django + PostgreSQL backend with Celery + RabbitMQ + Redis; JS / Node / React front end; Nginx + Docker deploy. Grew into tech-lead duties (code review, mentoring, roadmap input) as the team scaled to 3–5 engineers.
Built a real-time pipeline from London Stock Exchange WebSocket feeds → Elasticsearch → seq2seq LSTM trading model at tick-level (ns/ms) granularity. Sustained cross-server throughput with ingestion-rate-aware batching and backpressure isolation; ran the model head-to-head against the firm’s human traders over a multi-month evaluation window.
AngularJS front end for a Slack-like business messenger backed by Django REST. Owned the live-session lifecycle: state synchronization between client and server-authoritative state, socket-connection management with reconnect/backoff, presence/typing/read-receipts, and optimistic-UI vs. confirmed-state reconciliation. First professional web-dev role.
Education
3
Sep 2024 – Expected Dec 2026
Supervised by Dr. Mohammad Tayebi. Research on agentic AI systems for software security. CGPA 4.00/4.33; SFU Graduate Fellowship (Summer 2025). Selected courses: Secure Software Design (A+), Programming for Heterogeneous Computing Systems (A+), Directed Reading (A+).
Sep 2019 – Mar 2023
Shahid Bahonar University of Kerman
Thesis: Cryptocurrency Portfolio Management with Deep Learning — an LSTM + Q-learning agent integrating Twitter and news sentiment via a genetic algorithm. Supervisor: Dr. Hadis Mohseni. CGPA 18.72/20.
Sep 2013 – Aug 2019
Sharif University of Technology
Konkur national exam rank 813 of 250,000 (top 0.3%). Final project: Petopia, a smart pet-tracking IoT device with web and Android clients, in the Secure and Smart Systems Lab.
Awards
7 listed