Thanks to visit codestin.com
Credit goes to github.com

Skip to content

keyhankamyar/keyhankamyar

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

7 Commits
ย 
ย 

Repository files navigation

Hi, I'm Keyhan ๐Ÿ‘‹

Converting caffeine & curiosity into tomorrow's math

โ€œone gradient at a timeโ€

LinkedIn Badge Email Badge

๐Ÿ” Who I Am

# pip install curiosity==โˆž
from datetime import datetime

from pydantic import BaseModel


class Keyhan(BaseModel, frozen=True):
    """
    ML Engineer who treats research like spelunkingโ€”
    you don't know what you'll find until you're deep enough
    that turning back costs more than pushing forward.
    """

    power_on: datetime = datetime(1995, 1, 7)
    coords: tuple[float, float] = (32.6539, 51.6660)
    role: str = "ML Engineer & Researcher"
    fuel: str = "single-origin espresso, 18 g, 25 s, 93 ยฐC"

    principle: str = "Theory is the map; implementation is the territory."
    operating_mode: str = "10-12 hours/day, no weekends. It's not work; it's the game."

    focus: list[str] = [
        "Deep Reinforcement Learning (DRL) for complex control",
        "High-Performance Computing (HPC) & Custom Kernel Optimization",
        "Financial AI & Quantitative Systems",
        "Autonomous Agents & LLM Orchestration",
        "Scalable MLOps, Data Pipelines & Production Deployment",
        "Neural Architecture Search (NAS) & HPO",
    ]

    stack: dict[str, list[str]] = {
        "Languages": ["Python", "C++"],
        "ML Frameworks": ["PyTorch", "TensorFlow"],
        "RL Toolkits": ["Gymnasium", "RLlib", "Stable-Baselines3", "Tianshou", "CleanRL"],
        "HPO & NAS": ["Optuna", "Custom NAS Frameworks", "Ray Tune"],
        "Data Orchestration": ["Pandas", "NumPy", "Numba", "Parquet", "HDF5", "Apache Spark"],
        "Databases": ["MongoDB", "InfluxDB", "SQL"],
        "Deployment & MLOps": ["FastAPI", "Docker", "Git", "CI/CD", "Pydantic"],
        "Web Automation": ["BeautifulSoup", "Playwright"],
        "OS": ["Linux"],
    }

    def philosophy(self) -> str:
        return """
        The gap between a brilliant paper and a production-ready system is a chasm 
        of broken dependencies, hidden bottlenecks, and flawed assumptions. My work 
        lives in that chasm. I write the code that bridges itโ€”transforming theoretical 
        edge into tangible, performant, and maintainable software.
        """

๐Ÿงฐ My Stack

Category Tools
Languages Python C++
ML Frameworks PyTorch TensorFlow Lightning
RL Toolkits Gymnasium Stable Baselines3 Tianshou RLlib
LLM & AI OpenAI Ollama vLLM LangChain Llama.cpp OpenRouter
Data & Compute NumPy Pandas Numba Parquet Apache Spark CUDA
Databases MongoDB InfluxDB PostgreSQL
MLOps & Optimization Optuna Ray W&B Pydantic
Deployment FastAPI Docker Git Linux
Web & Automation Playwright BeautifulSoup AsyncIO

๐Ÿ•ธ๏ธ Projects Web & Roadmap

This chart shows how my projects build upon and connect with each other, forming a cohesive research and development ecosystem.

%%{init: {'theme':'base', 'themeVariables': { 'primaryColor':'#667eea','primaryTextColor':'#fff','primaryBorderColor':'#7C4DFF','lineColor':'#F8B229','secondaryColor':'#764ba2','tertiaryColor':'#1e3a8a','background':'#0f172a','mainBkg':'#1e293b','secondaryBkg':'#334155','tertiaryBkg':'#475569','textColor':'#f1f5f9','fontSize':'16px','fontFamily':'ui-monospace'}}}%%
graph TD
    subgraph CoreLib["๐Ÿ”ง Core Libraries"]
        A[Proxy Rotator]:::shipped
        B[Tick Vault]:::shipped
        C[SpaX]:::shipped
    end

    subgraph Frameworks["โšก Frameworks & Guides"]
        D[Lightning HPO Playbooks]:::polishing
        E[LLM Benchmarking]:::planned
    end

    subgraph Applications["๐Ÿš€ Application & Research"]
        F[Financial Env]:::planned
        G[CV/CL Agent]:::planned
        H[Clean TS]:::planned
    end

    C -.->|Config & HPO| D
    C -.->|Config & HPO| H
    C -.->|Experiment Design| F

    D -.->|Training & Eval| H
    D -.->|Training & Eval| F

    E -.->|Model Selection| G

    A ==>|Resilience| G
    B ==>|Data Source| F
    
    classDef shipped fill:#4CAF50,stroke:#2E7D32,stroke-width:3px,color:#fff,font-weight:bold
    classDef polishing fill:#FFC107,stroke:#F57F17,stroke-width:3px,color:#000,font-weight:bold
    classDef planned fill:#64748b,stroke:#475569,stroke-width:2px,color:#fff
    
    style CoreLib fill:#1e3a8a,stroke:#3b82f6,stroke-width:3px,color:#fff,stroke-dasharray: 5 5
    style Frameworks fill:#7c2d12,stroke:#f97316,stroke-width:3px,color:#fff,stroke-dasharray: 5 5
    style Applications fill:#581c87,stroke:#a855f7,stroke-width:3px,color:#fff,stroke-dasharray: 5 5
Loading
Legend: ๐ŸŸข Shipped โ€ข ๐ŸŸก Polishing โ€ข โšซ Planned

๐Ÿ“‹ Project Details

# Project Description Status
1 Proxy Rotator A production-ready library that seamlessly integrates proxy rotation into httpx clients (sync/async). Built for resilience and simplicity in web scraping and API automation workflows. ๐Ÿš€
Shipped
2 Tick Vault High-fidelity financial tick data scraper for Dukascopy Bank (Swiss). Extracts raw, sub-second precision market data for quantitative analysis and backtesting. ๐Ÿš€
Shipped
3 SpaX Pythonic, type-safe search space configuration for HPO, NAS, and ML experiment tracking. Declarative configs with conditional parameters, automatic validation, and zero boilerplate. Pydantic-based with native Optuna integration. ๐Ÿš€
Shipped
4 Lightning HPO Playbooks Industry-standard examples and guides for model training, optimization, and research using PyTorch Lightning. Covers SOTA practices for NAS, HPO, distributed training, and production-ready ML pipelines. ๐Ÿ”จ
Finishing Touches
5 CV+CL Agent An agentic framework that auto-generates tailored CVs and cover letters optimized for specific job postings. Combines LLM orchestration with structured outputs for efficient application workflows. โณ
Up Next
6 Financial Env A blazingly fast, parallelized Gymnasium environment for algorithmic trading. Includes a placeholder reward functionโ€”the real one stays private (years of research aren't free). Built for large-scale RL training; production readiness TBD pending capital for full-scale experiments. ๐Ÿ“
Planned
7 LLM Benchmarking A practical guide for building use-case-specific LLM evaluation pipelines. Generic benchmarks mislead; this teaches how to design reliable, domain-aware benchmarks that actually reflect real-world performance. ๐Ÿ“
Planned
8 Clean-TS A modular, Pythonic reimplementation of canonical time-series architectures. Traditional TS codebases are archaic and opaqueโ€”this makes them readable, extensible, and reproducible. Requires ~1 month of polish before release. ๐Ÿ“
Planned

ย ย 

ยฉ 2025 Keyhan Kamyar ยท Fueled by espresso and curiosity ยท Built with ๐Ÿ’œ and late nights