Work

Research in a lab, and eight years of ML infrastructure that had to stay up at night.

  • Nov 2025 — present

    Doctoral Researcher

    Elios Lab, DITEN — Università degli Studi di Genova, Italy

    • Hardware-aware benchmark of fifteen single and hybrid visual-backbone configurations for real-time driving-scenario video understanding, on H100 and Jetson Orin Nano.
    • Quantified FP32/FP16/INT8 precision trade-offs on directly metered hardware — throughput, accuracy and energy per frame.
    • Released the benchmark harness publicly at Elios-Lab/video-encoder-benchmark.
    • Work conducted with a cross-institution group spanning UniGe, QMUL and STMicroelectronics.

  • Apr 2017 — Jun 2025

    Snapp

    Tehran, Iran — Iran's largest ride-hailing platform · 50M+ registered users · ~4M rides/day at peak

    • Senior Engineering Manager — Backend & AI Platform2023 — 2025
    • Engineering Manager — AI / ML Platform2022 — 2023
    • Technical Specialist — ML & Backend2019 — 2022
    • Senior Software Engineer — Backend2017 — 2019
    • Founded and led an 11-engineer AI Platform organisation (5 platform, 6 backend) with architecture ownership and 24/7 production on-call.
    • Architected the online/offline real-time feature store serving ~100M inferences a day at sub-50 ms p99, including parity checks between offline training and online serving features to prevent silent training/serving skew.
    • Owned the A/B programme for the AI dispatching model — experiment design, staged ramp, rollback criteria — and measured a ~2% lift in completed rides against holdout, roughly 100k additional completed rides a day.
    • Standardised a Python/Go model-serving pattern with low-latency inference gateways as the platform's default deployment path for every new ML use case.
    • Built MLOps CI/CD for model promotion: versioning, validation gates, staged rollout, drift and skew monitoring, automated alerting.
    • Built the company's governed LLM access layer on LiteLLM and Langfuse — unified model access, per-team virtual keys and rate limits, request tracing, cost attribution.
    • Migrated ~5 high-traffic PHP/Laravel services to event-driven Go on NATS, holding p99 steady through roughly 10× user growth; raised legacy test coverage from 20% to 87%.
    • Grew the backend organisation from 4 to 30 engineers in six months as technical interviewer and hiring lead.

  • Sep 2016 — Apr 2017

    Senior Software Engineer — Backend

    GSM Group — Tehran, Iran

    Backend and data services for telecom products in Django, PHP and MySQL. Query tuning, indexing and a Redis cache layer cut execution time on high-frequency APIs by about 60%; integrated third-party data-ingestion pipelines for three internal teams.

Teaching & mentoring

  • 2026

    Instructor — Agentic AI Mini-Camp

    Quera

    Project-based course on building AI agents rather than prompting models: RAG and vector databases, orchestration with LangChain / Flowise / n8n, tool design, memory, and human-in-the-loop workflows.

  • 2026 — present

    Mentor

    ADPList

    Free 1:1 sessions on backend and ML system design, Kubernetes and model serving, and career direction for engineers moving toward senior and lead roles.

  • 2019 — 2025

    Internal instructor & mentor

    Snapp

    Taught internal courses on MLOps (2021) and Machine Learning Fundamentals (2019); hired, onboarded and mentored 20+ backend and ML engineers from screening through to autonomous production ownership.

  • 2010 — 2016

    Teaching Assistant

    Islamic Azad University — South Tehran and Kashan branches

    Advanced Databases; Advanced Topics in Software Engineering; Database and Operating Systems workshops; Algorithms; Language and Machine Theory.

Tools I work with
Languages
Python · Go · Rust · SQL · Bash
ML & serving
PyTorch · TensorFlow/Keras · scikit-learn · XGBoost · ONNX · vLLM · SGLang · llama.cpp · Triton · Hugging Face · LoRA/QLoRA · quantisation
LLMOps
LiteLLM gateway · Langfuse tracing · RAG and GraphRAG · multi-agent orchestration · MCP
Platform
Kubernetes · Docker · Helm · Terraform · Argo CD · GitLab CI · GitHub Actions · OpenShift · AWS
Data
Kafka · NATS · Airflow · MLflow · Weights & Biases · Spark · Flink · ClickHouse · Redis · PostgreSQL
Observability
Prometheus · Grafana · OpenTelemetry · drift and training/serving-skew monitoring
Hardware
NVIDIA H100 · Jetson AGX Orin / Orin Nano · Raspberry Pi 5 · STM32MP2 · GPU profiling