Raihan Jashil
The Complete Record Résumé · PDF

The one who reads what has not yet been written. Who finds the signal sleeping in noise, who parts the grey fog with numbers, who sees the shape of what is coming. Above the uncertainty, he keeps the pattern.

Raihan Jashil

AI Engineer

About The One Who Reads the Fog

It started with a question I couldn't put down. How did we get clever enough to teach stone to think, and could I understand it well enough to do it myself? That curiosity has more or less run my life since I was a kid.

It's why I took computer science in high school, and why I kept exploring once I got to university. My first two years at UDST let me get my hands on a bit of everything: frontend, backend, databases, a little security. I wanted to see how the whole machine fit together before I picked the corner of it I'd make mine. Then I took intro to data science and AI, and the question I'd been chasing since I was a kid turned into the thing I actually got to build. I switched into it right away and I've been exploring it ever since.

I go all the way down on things I care about. I've 100%'d the games I love, and I work the same way, with nothing left unfinished. What pulls me most right now is building AI tools that hold up once they're in someone's hands, and the research that makes better ones possible.

Experience The Sequence

  1. Sequence 06

    The Reader of the Great Works

    Sent to establish what is already true. To read the great existing models of the genome and rebuild them stone by stone, testing whether their claims hold before any new work is measured against them.

    Research Intern Qatar Computing Research Institute (QCRI)

    Ar-Rayyan, Qatar · May 2026 – Present

    • Baselines lead on a deep learning research project for genomic mutation effect prediction. Evaluated four genomic foundation models (Enformer, Borzoi, Nucleotide Transformer, AlphaGenome) and mapped their transformer architectures to inference requirements.
    • Engineered end-to-end Python inference pipelines for large transformer models on hg38 reference sequences up to 524,288 bp, implementing sequence extraction, center-padding logic, and output validation against experimental ground truth via Pearson correlation across 5,313 genomic assay tracks.
    • Containerized per-model environments with reproducible conda setups, resolved cross-platform dependency conflicts, and deployed models to remote GPU infrastructure over SSH.
  2. Sequence 07

    The Diviner in the Halls of Healing

    Took the tangled records of a nation's care and gave them a voice its leaders could act on.

    Data Science Intern Primary Health Care Corporation (PHCC)

    Doha, Qatar · January 2026 – March 2026

    • Recognized by the Head of Business and Health Intelligence for high-impact work, earning a formal referral for post-graduation employment, with two dashboards considered for departmental publication.
    • Built Power BI dashboards using DAX for senior management and operations teams, translating complex health data into actionable insight.
    • Cleaned and preprocessed multi-source datasets and built robust ETL pipelines to ensure data quality and integrity for modeling.
  3. Sequence 08

    The One Who Taught Machines to Read

    Built constructs that read what people found tedious, turning stacks of unreadable paper into order.

    AI/ML Freelance PrimeTax

    Abu Dhabi, UAE (Remote) · June 2025 – December 2025

    • Built two production web applications for automated invoice data extraction, integrating the Google Gemini API for LLM-based document understanding.
    • Engineered a multi-format document-processing pipeline (PDF, JPG, PNG, TIFF, BMP, WEBP) with intelligent 5-page PDF chunking and dynamic line-item detection to handle varying invoice layouts.
    • Automated structured Excel report generation with custom formatting, branding, and file-splitting features.
  4. Sequence 09

    The First Step Beyond

    Where the path began. The first models built, the first results carried before those who knew more.

    AI/ML Intern Vismaya Infotech Solutions

    Doha, Qatar · June 2024 – August 2024

    • Designed and optimized a threshold-based classification model, reaching 86% accuracy, and presented results to technical leadership.
    • Cleaned and preprocessed multi-source datasets and built pipelines to ensure data quality and integrity for modeling.

The climb is not finished. The next seat waits above.

Projects Sealed Artifacts

A bilingual Arabic and English calligraphy tutor built on Fanar, Qatar's LLM.

  • Architected a strict split where deterministic code handled trajectory geometry, scoring, and workflow control, while Fanar models (Fanar-C-2-27B for feedback, Oryx for vision, Shaheen-MT for translation) handled language, pedagogy, and visual assessment.
  • Orchestrated multi-step agentic workflows with LangGraph, including a curriculum generator with validate/repair loops and a sentiment-to-calligraphy pipeline, with Guard-gated generation, Zod-validated model I/O, and deterministic fallbacks that keep output coherent even with the LLM fully offline.
  • Delivered a full-stack Next.js / React 19 / TypeScript app with Supabase authentication, Postgres, row-level security, and private storage, using Fanar Oryx to segment and score photographed paper worksheets against digital references.

Impact: HighRole: BuilderRecognition: 2nd Place, Fanar Hackathon (QCRI), June 2026

An end-to-end AI content studio on a FastAPI backend.

  • Integrated OpenAI GPT-4o to score drafts across Twitter/X, LinkedIn, and Facebook on a five-dimension rubric (hook, clarity, CTA, format, tone), with platform-specific hard-rule checks, per-persona rewrites, and bilingual English and Arabic detection.
  • Engineered an AI video pipeline using faster-whisper transcription, scene analysis, and natural-language chat-driven editing, plus an FFmpeg music mixer with three modes (background blend with speech-aware auto-ducking, full replace, keep-original) driven by GPT mood detection over Jamendo tracks and ffprobe audio analysis.
  • Implemented OAuth 2.0 with PKCE for account connection and direct publishing, versioned non-destructive renders, and a resilient architecture where the optional heavy media stack degrades gracefully to HTTP 503 without blocking the core API.

Impact: HighRole: BuilderRecognition: 2nd Place, QSTP Hackathon (Stars of Science Edition), June 2026

An AI-driven mobile banking platform for financial literacy.

  • Designed and served an MLP neural network reaching 91% accuracy in financial-profile classification, benchmarked against Random Forest (85%) and XGBoost (83.5%) across seven algorithms.
  • Built a feature-engineering pipeline over 16 financial metrics with SMOTE for class imbalance, powering an adaptive learning system with age-stratified content and automated struggle detection.
  • Deployed the model in production via FastAPI microservices with real-time inference, a Firebase/MongoDB backend, and integrated model versioning for continuous deployment.

Impact: HighRole: AI/Data Science LeadContext: Presented at the Qatar Fintech Hackathon, May 2025

Recovered fragment

Recovered fragment. Over the span of three days, the subject taught himself the arts of 3D modeling in Blender and the modding of games, assembling a small network of strangers to help, and bound the likeness of a close friend into a construct that he then loosed inside a game as a mod.

The intent was never malice. It was mischief, aimed precisely at the one person who would understand it. The friend's horror gave way to laughter, which was the only outcome the subject had ever intended.

The artifact was purged the moment its purpose was served. What remains is the proof of what it took to make: three days, from knowing nothing to a working construct, learned from scratch and shipped, for a joke.

Assessment. Learning velocity: extreme. Initiative: unprompted. Judgment: intact, seal honored, artifact destroyed.

Achievements Records of Renown

  • Huawei AI Innovation National Competition

    Engineered a neural network for financial habit detection in youth demographics, reaching 89% prediction accuracy.

    4th place nationally Fall 2024
  • Samsung Innovation Campus, Qatar

    Developed a CNN for food classification on large-scale Kaggle datasets as part of Qatar's first Samsung Innovation Campus cohort, reaching an 86% recognition score.

    Fall 2023 – Spring 2024
  • Qatar Collegiate Programming Contest (CMUQ)

    Competed in algorithmic programming contests, applying data structures, algorithms, and optimization techniques under time constraints.

    Fall 2024 – Fall 2025
  • Fanar Hackathon (QCRI)

    See Sealed Artifact 001, "Midad."

    2nd place June 2026
  • QSTP Hackathon (Stars of Science Edition)

    See Sealed Artifact 002, "PostPulse."

    2nd place June 2026

Skills The Grimoire

Tongues Languages

  • Python
  • SQL
  • JavaScript
  • HTML
  • CSS

Rituals and Frameworks Libraries and Frameworks

  • PyTorch
  • TensorFlow
  • scikit-learn
  • XGBoost
  • pandas
  • NumPy
  • FastAPI
  • LangGraph
  • Matplotlib
  • Seaborn

Instruments Tools and Databases

  • MongoDB
  • PostgreSQL
  • Firebase
  • Google Gemini API
  • Git
  • Power BI
  • DAX
  • QlikView

Recognition Witnesses

Recognized by Mr. Chiragkumar Gohel, Head of Business and Health Intelligence at PHCC, for delivering high-impact work, resulting in a formal referral for post-graduation employment, with two dashboards considered for departmental publication.

Contact Offer Tribute

Cross the fog to reach me. Leave your tribute below.

Currently seeking roles in applied AI and research.

This sends a message directly to Raihan.