Skip to content
SINA
Let's Chat

Hi, I am Sina

Softwareengineer,deepinAI,buildingsystemsthatsurviverealdata,realusersandproduction.

Flowing lines of light converging into a network of nodes
Range

Applied AI for Every Workflow

Pharma distribution, healthcare support, venture capital, maritime maintenance, B2B market data. The domains change, the job doesn't: find the expensive manual process, measure it honestly and replace it with something a team can trust.

See all projects
Retrieve
  • Maritime
  • Sports & Education

  • Software Engineering First

    Model design through APIs, databases, tests and deployment, because a model nobody can run isn't a product.

  • Evidence Over Vibes

    Baselines first, evaluation suites always, and a source you can click for every claim a system makes.

  • Built for the Real Constraints

    Sparse data, Persian users, strict business rules and networks that fail in interesting ways.

Portfolio

Selected Projects

  • 01 / Healthcare AI / AI Software Engineer

    Medical RAG Chatbot

    • RAG
    • Hybrid Search
    • Reranking
    • Persian NLP
    Read Case Study
    Diagram of a question flowing through keyword and vector search into a reranker, then a cited answer or a human escalation

    A production bilingual support assistant for Spartina. It answers Persian patient questions from a curated English knowledge base and escalates complex cases to people.

  • 02 / Optimization / AI Software Engineer

    Workforce Scheduling Optimizer

    • OR-Tools
    • CP-SAT
    • Python
    • Fairness
    Read Case Study
    A monthly roster grid being solved cell by cell, with zero violations and zero unfilled slots

    Constraint programming for a 35-person healthcare support operation. Monthly planning went from up to 8 days to under 2 hours, with zero violations and no unfilled slots.

  • 03 / Machine Learning / AI Software Engineer

    Sales Target Allocation

    • Hurdle Model
    • Gradient Boosting
    • Sparse Demand
    Read Case Study
    A monthly product target split through a two-stage model into many pharmacy allocations that sum back to the target

    A two-stage hurdle model that distributes pharmaceutical sales targets down to individual pharmacies, then reconciles every prediction against strict monthly product totals.

  • 04 / Document AI / Author

    Deal Flow Automation

    • Local-First
    • LLMs
    • FastAPI
    • Next.js
    Read Case Study
    A pitch deck being scanned into extracted facts with evidence links and a scorecard

    A document intelligence platform that turns unstructured pitch decks into evidence-backed investment scorecards, without confidential files ever leaving the machine.

  • 05 / AI Agents / Technical Lead

    Marja Intelligence

    • Agents
    • Tool Calling
    • SQLite FTS5
    • RTL
    Read Case Study
    A Persian chat interface where the assistant answers with a table of companies built from a database

    A Persian chat assistant for B2B market research over 15,913 companies. Result tables are always built by the server from SQL, so the model can't invent a company.

  • 06 / Full Stack / Technical Lead

    Marja Panel

    • Django
    • Next.js
    • PostgreSQL
    • Live Product
    Read Case Study
    Screenshot of the Marja Panel home page with its search bar and industry categories

    A subscription platform for Iranian company data, with a metered quota engine, search and an admin data programme. Live at marjapanel.ir.

Research

Selected Research & Experiments

01/04

Hand Pose
few-shot learning

A 3D hand skeleton with faint alternative poses around it
Ridgelines of a time series with forecasts arriving from both directions
What I Reach For

The Core
Toolkit.

  • LangChain

  • LangGraph

  • spaCy

  • Weaviate

  • ChromaDB

  • FAISS

  • RAGAS

Timeline

How I Got Here

From teaching gifted students to shipping AI into operations.

  1. Co-founded BABAEEE

    2020 – 2024Non-Profit

    A non-profit learning platform for students preparing for National Olympiads. I led front-end development, then moved into HR, recruiting and sponsorships.

  2. University of Tehran

    2020 – 2025B.Sc. Computer Science

    Information retrieval, data mining, bio-inspired computing, AI and a 20/20 in Advanced Programming.

    ThesisSep 2025

    Emerging few-shot learning techniques for hand pose estimation, bridging theory and practice.

  3. Rahnema College

    Mar – Jun 2024Data Analytics Bootcamp

    170+ hours of statistical inference, BI, A/B testing and data storytelling on real-world datasets.

  4. Sharif University

    Jan – Apr 2025AI Researcher

    Research under Dr. Peyvandi combining forecasting and backcasting, from statistical methods to diffusion-based deep models.

  5. Orchid Pharmed

    Jun 2025 – PresentAI Software Engineer

    Production AI and operations research across pharmaceutical distribution and support: LLMs, constraint programming and predictive analytics.

  6. Built Maham Academy

    2026Full Stack

    A full registration, payments and attendance platform for a basketball academy, with CI that deploys from merge to server.

  7. Marja Goes Live

    Sep 2026Launch

    A subscription platform for Iranian company data reaches production at marjapanel.ir, with Marja Intelligence built beside it.

Scroll to explore

Latest Writing

Notes from shipping AI the hard way.

Browse all writing
  1. 01Forecasting

    When predicting zero beats every model you have

    Read More
  2. 02AI Agents

    Grounding by construction: never let the model write the table

    Read More
  3. 03Optimization

    "Soft" doesn't mean optional: lessons from scheduling real people

    Read More
  4. 04Engineering

    Shipping from Iran: when the network is part of your architecture

    Read More
How I Work

Measured, shipped
and owned

I like hard, slightly messy problems where a real team is losing real time. I'll tell you early if a rule would do better than a model, and I stay until the thing runs in production.

  • Not demos that die in a notebook.
  • Not accuracy without a baseline.
  • Not AI where a rule would do.
  1. 01

    Operations Teams

    Drowning in spreadsheets, rosters and manual allocation

  2. 02

    Founders

    Who need an AI product in production, not a slide about one

  3. 03

    Regulated Domains

    Where every answer needs a source and an audit trail

  4. 04

    Data-Heavy Businesses

    Sitting on documents and records nobody can query

  5. 05

    Research Groups

    With ideas worth testing rigorously, end to end

Got a messy problem worth automating?

Send me the ugly version. It's usually the interesting one.

Let's talk

Collaborations

Worked with

  • Orchid Pharmed
  • Spartina
  • Marja
  • Apex Labs
  • Maham Academy
  • University of Tehran
  • Sharif University of Technology
  • Rahnema College
  • BABAEEE

& a few teams who'd rather stay anonymous