Real problems. Connected components. Explore each case study without leaving this page.
AI & Data / 01
▤SQL / CSV / XLSX
◇Schema + embeddings
⌘SQL generation
↗Validation + execution
Natural language → structured insight
SmartEDA
An AI data platform that turns questions in Persian into SQL queries and understandable answers.
DjangoPostgreSQLCeleryRedisLLMsCasdoor
Explore the engineering ＋
My contribution
Worked on backend and AI integration for the data-analysis platform at Sharif University.
Problem
Make SQL databases, CSV files and Excel workbooks accessible through natural-language questions, including Persian.
Architecture
A Django REST API manages data sources and conversation history. Celery and Redis support asynchronous training. Schema context and embeddings support LLM-driven SQL generation, while a React/Vite frontend connects through Casdoor SSO.
Engineering challenges
Extract usable SQL from noisy model output.
Represent schemas across database and file sources.
Coordinate background training with conversational queries.
Integrate SPA authentication and production deployment.
Decisions & approach
Keep ingestion and training in asynchronous workflows.
Treat SQL extraction, validation and execution as distinct stages.
Maintain source context alongside chats and messages.
Outcome
A platform for connecting structured sources, asking questions and receiving query results with Persian explanations.
Infrastructure / 02
▤Admin interface
◇JWT + permissions
⌘Service layer
↗Casdoor / LiteLLM
One interface. Multiple services.
Superadmin
Centralized administration for identity services and AI infrastructure, with explicit service boundaries.
Django 6DRFCasdoorLiteLLMDockerPostgreSQL
Explore the engineering ＋
My contribution
Worked on the Django backend and integrations with Casdoor and LiteLLM.
Problem
Bring administration of multiple services into a consistent backend interface.
Architecture
Django REST Framework exposes administration endpoints through service layers and external API clients. PostgreSQL stores application data, with Redis and Celery supporting background work. Docker and GitHub Actions support deployment workflows.
Engineering challenges
Model different external service APIs consistently.
Integrate identity administration across users, organizations and permissions.
Manage environments and deployment configuration.
Decisions & approach
Separate external API clients from application service logic.
Group identity capabilities around explicit resources.
Use container-based deployment and CI for repeatable workflows.
Outcome
An internal administration platform integrating identity operations and AI infrastructure management.
Backend / 03
▤Phone + OTP
◇Synapse login token
⌘Bots + dynamic forms
↗Submission workflow
Communication meets structured workflows
Agrino
A Matrix-based communication platform connecting phone authentication, bots and conditional forms.
PythonDjangoMatrixSynapseOTP
Explore the engineering ＋
My contribution
Worked on Matrix/Synapse integrations, authentication and Django-backed workflow functionality.
Problem
Connect real-time communication with structured submissions and phone-based user access.
Architecture
Django integrates with Matrix/Synapse endpoints for authentication, forms and bot workflows. Questions support typed fields and dependencies; submission states express draft, submission, review and follow-up.
Engineering challenges
Coordinate OTP verification and login-token flows.
Handle rate limits, multipart requests and bot sessions.
Represent conditional questions and submission states.
Decisions & approach
Represent question types and dependencies explicitly.
Use submission states to express review and follow-up.
Integrate phone verification with Matrix login-token flows.
Outcome
Communication and backend workflow functionality spanning authentication, dynamic forms, bots and submissions.
Data / 04
▤Technology records
◇Filters + sectors
⌘Analytics API
↗Radar + dashboards
A structured view of emerging technology
Technology Radar
A visualization-oriented backend for exploring technologies by impact, timeline and sector.
DjangoDRFReactAnalyticsPermissions
Explore the engineering ＋
My contribution
Developed/worked on the Django REST backend for the technology radar platform.
Problem
Organize technology information so users can filter, compare and evaluate its impact and outlook.
Architecture
Structured technology records connect to sectors, filters and subfilters. DRF endpoints expose impact and timeline analytics for a React interface, with permission checks for restricted business filters.
Engineering challenges
Model technology information and nested filters.
Provide useful aggregate metrics and range buckets.
Restrict business filters by user permissions.
Decisions & approach
Expose aggregate impact and timeline data through analytics APIs.
Model access rules alongside filter capabilities.
Organize overview, challenges, outlook and resources as structured content.
Outcome
A platform for filtering technology information and exploring impact, timelines and analytics.
02 /
Where I’ve been building
Jul 2025 — Present
Sharif University
Software Engineer · AI Platform
Developing AI/data-analysis platforms around enterprise data: file ingestion, schema understanding, asynchronous training and Persian natural-language-to-SQL.
DjangoLLMsPostgreSQLCelery
Apr 2024 — Apr 2026
Nona Startup Studio
Software Engineer
Developed event-driven backends, third-party integrations and real-time chat. Built Docker deployment automation across three environments and worked on barcode recognition, Logistics Radar and Optipack.
DRFDockerFastAPIYOLO
Jun 2023 — Mar 2024
Actin
Software Developer — Data Focus
Designed data architecture for a Smart Mirror product and assisted development of skeleton anomaly detection with OpenCV.
PythonOpenCVData architecture
Nov 2022 — May 2023
Shanbeh Mag
Software Developer
Developed an Airflow data pipeline for crawling registered users’ LinkedIn accounts, alongside Django REST chat functionality.
AirflowETLDRF
Apr 2022 — Oct 2022
ToseKara
Back-end Developer
Developed core insurance software, designed and implemented microservices, and built a passport-reader application with Django and OpenCV.
DjangoMicroservicesOpenCV
Earlier: Dorna Wings
Earlier / internship work with Python, Arduino, MAVLink and OpenCV for drone control, position holding and hardware testing.
03 /
Curiosity, in practice.
Public repositories, smaller projects and active experiments. Each has its own scope.
PUBLIC REPOSITORY / IoT
Smart Greenhouse
Sensor-connected plant care with Arduino, Raspberry Pi and a web service.
PythonArduinoRaspberry PiGPIOHTTP
How it works ＋
Arduino reads light and soil-moisture sensors; Python on Raspberry Pi combines serial readings with DHT11 temperature and humidity. HTTP scripts send sensor data and request device states, while GPIO code controls pump and lamp relays.
Scope of the experiment
A hardware integration experiment. The repository contains older module-name inconsistencies and is not presented as a ready-to-deploy controller.
An image-classification experiment connecting a Keras CNN to OpenCV camera capture.
PythonTensorFlow / KerasOpenCVCNN
How it works ＋
The code defines convolution and max-pooling layers followed by a softmax classifier. Images are resized to 64 × 64; preprocessing uses shear, zoom and horizontal-flip augmentation. A saved model and class mapping support predictions from a captured image.
Scope of the experiment
An experimental pipeline, not a validated biometric system. Training and validation generators use the same source directory; no independent accuracy result is claimed.
Video processing with Haar-cascade car detection and HSV-based color segmentation.
PythonOpenCVNumPyHaar cascadeHSV
How it works ＋
OpenCV reads video frames and detects cars with a Haar cascade. Separate red, green and blue HSV masks are dilated, converted to contours and drawn as bounding boxes when their area exceeds a threshold.
Scope of the experiment
The color masks run across the entire frame, not within each detected car. Some displayed color labels differ from their masks, so this is presented as a detection experiment rather than a validated vehicle-color classifier.
LiteLLM with OpenTelemetry, Jaeger and PostgreSQL. Exploring model routing, end-user attribution, usage tracking and billing limits.
LiteLLMOpenTelemetryJaeger
◌ Infrastructure exploration
AI Agents
Agents with state
Exploring LangGraph, tool calling, structured responses, asynchronous invocation and context injection for AI applications.
LangGraphLangChainPython
◌ Active study & experiments
AI Agents
Persian NL-to-SQL
Retrieve schema context, generate SQL, execute a query and explain the results in Persian.
SQLAlchemyRedisEmbeddings
◌ Prototype
Computer Vision
Barcode recognition
A barcode-reading and recognition application combining an API service with computer vision.
FastAPITensorFlowYOLO
◌ Developed at Nona
Computer Vision
Virtual watch try-on
Exploring a Python AI service for WooCommerce, with particular attention to preserving physical watch scale across different wrist sizes.
PythonWooCommerceCelery
◌ Architecture & prompting experiment
AI Agents
Software-generation agents
Investigating agents that plan applications, generate code, execute in containers, debug and preview results.
AgentsContainersPython
◌ Exploration
04 /
I like knowing how the pieces fit.
Python and backend development are my foundation. Over more than four years of professional work, my scope has expanded into data pipelines, asynchronous workflows, authentication, deployment and AI applications.
Working with structured data led me toward natural-language-to-SQL. Integrating LLMs brought questions about provider management, request tracing, context and how model output fits into an application.
I’m growing toward stronger backend and systems engineering for AI-native products. IoT, robotics and computer vision remain long-running interests, from drones to sensor-connected hardware.
I care about the whole application: its data model, access rules, background work and behavior in production.