I am a Statistics student at the University of Tehran building reliable machine learning and applied AI systems. My work sits at the intersection of statistical reasoning, software engineering, and practical decision support.
I care about the parts of AI that make systems useful in the real world: calibrated uncertainty, evaluation under distribution shift, explicit data boundaries, safe tool access, and interfaces that let people inspect and challenge a result.
I previously studied Electrical Engineering and completed Maktab Sharif's software-development program. I use AI tools throughout development for implementation, testing, debugging, and documentation, while checking important decisions against code, data, and reproducible artifacts.
| Project | What I built |
|---|---|
| PromoGuard Retail Intelligence | An end-to-end decision-support system for evidence-aware promotion auditing on real retail data, combining time-aware forecast comparison, uncertainty guardrails, explicit abstention, FastAPI contracts, and a Persian Streamlit review interface. |
| Calibrated Predictive Reliability | A completed, reproducible C-MAPSS study of remaining-useful-life prediction under operating-condition and fault-mode shift, with calibration analysis, leakage-safe splits, frozen protocols, immutable artifacts, and independent reconstruction. |
| AURALIS | A full-stack Persian and multilingual meeting-intelligence product with speech capture, evidence-grounded insights, workspaces, action tracking, and a versioned release history. |
| Bearing Prognostics & Value of Information | An in-progress research pipeline for sequential bearing-degradation detection, false-alarm control, and inspection decisions, with separate development, calibration, and external NASA IMS validation. |
| Professor-Aware Exam Coach | A local-first study workspace with course-source retrieval, structured feedback, FastAPI, Next.js, SQLite, and explicit human-review boundaries. |
- Tenuo #626: added a tested Next.js App Router Node-runtime deployment example using the packed
@tenuo/coretarball. - randkv #12: verified the Transformers adapter against
HuggingFaceTB/SmolLM2-135Mand documented model-family compatibility.
- Secure AI Gateway: a small, credential-free MVP for exploring prompt-injection checks, least-privilege tool access, rate limiting, and privacy-aware audit logs. It is an experiment and learning artifact rather than a production security product.
I define the claim before running an experiment, keep data boundaries explicit, and document what the available evidence does and does not support. I prefer small, inspectable systems with clear tests and reproducible artifacts over opaque demos.
My current stack includes Python, scikit-learn, FastAPI, TypeScript, React, Next.js, SQLite, Streamlit, GitHub Actions, and LLM/RAG tooling. I choose the stack around the problem rather than treating the stack as the point of the project.
My current direction is reliable decision-making for sequential and structured data, together with safe AI infrastructure: calibrated prediction, evaluation under shift, uncertainty-aware policies, and guardrails for LLM applications that use tools or sensitive data.
For research, software, or data/AI product collaboration, reach out through GitHub or LinkedIn.