Skip to content
View mdumr4's full-sized avatar

Highlights

  • Pro

Block or report mdumr4

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mdumr4/README.md

⚡ Executive Summary & Engineering DNA

engineer:
  name: "Mohammed Umar F"
  title: "AI Systems Architect & Multimodal ML Engineer"
  core_philosophy: "Production-first intelligence — bridging scalable data architectures with frontier neural transformers."
  specializations:
    - multimodal_deep_learning: "Vision-Language fusion (DINOv2, ResNet34, DistilBERT, cross-attention)"
    - information_retrieval: "Two-stage hybrid RAG (FAISS dense + SPLADE-v3 sparse + Cross-Encoders)"
    - entity_resolution: "Million-scale record linkage with TF-IDF n-gram inverted indexing & Levenshtein"
    - autonomous_agents: "Stealth Chrome MV3 DOM automation, GPT-4o-mini lead qualification & Supabase"
  current_quest: "Scaling production multimodal inference pipelines, RAG systems, and enterprise agent workflows"

🌟 Featured Production Systems & Architecture Showcase

System / Repository Domain & Focus Core Architecture & Tech Stack Key Metric
📄 resume-ranker
Enterprise Two-Stage Hybrid RAG
Information Retrieval 		Two-stage hybrid candidate retrieval: dense FAISS vector search fused with sparse SPLADE-v3 expansions, two-part Cross-Encoder reranking, and Qwen 2.5 synthesis. 100% explainable candidate rankings
🏷️ Smart-Product-Pricing
Enterprise Multimodal Pricing Engine
Amazon ML Challenge 		Dual-tower architecture fusing DistilBERT text embeddings with DINOv2 visual features via cross-attention, fed into gradient boosted LightGBM with Snappy Parquet streaming. 0.38 RMSLE benchmark
🌿 PlantDocBot
Multimodal Agronomy Diagnosis
Infosys Springboard 		Dual-tower agronomy diagnosis platform fusing ResNet34 foliage visual classification with fine-tuned DistilBERT symptom extraction, dynamic confidence gating, and Streamlit UI. 95.2% diagnostic accuracy
🤖 Eylo-Reddit-v2
Autonomous AI Growth & Outreach
Agentic Systems 		Manifest V3 Chrome extension featuring anti-detection DOM humanizers, GPT-4o-mini intent qualification, and unified Supabase multi-account lead deduplication. 10x qualified lead discovery
🏢 Business-Entity-Resolution
Million-Scale Merchant Deduplication
Data Engineering 		Scalable record linkage pipeline combining TF-IDF n-gram inverted index blocking, Levenshtein distance matrices, and a gradient boosted classification reranker. > 99% candidate pair recall
🥗 nutritional-ai-agent
Autonomous Clinical Dietary Reasoner
IBM Hackathon 		Autonomous multi-agent dietary assistant orchestrating biomarker ingestion, clinical constraint solvers, and adaptive recipe synthesis for personalized medical nutrition. Zero contraindicated dietary plans

🛠️ Technical Arsenal & Engineering Competencies

💻 Core Languages & Runtimes

🧠 Deep Learning, Computer Vision & Machine Learning

🔍 Information Retrieval, RAG & LLMs

⚙️ Automation, Backend & Cloud Infrastructure

📊 GitHub Activity & Real-Time Performance Matrix

🏆 Hackathon & Engineering Track Record

  • 🥇 Amazon ML Challenge: Engineered a dual-tower multimodal pricing pipeline merging text transformers, self-supervised DINOv2 vision encoders, and cross-attention fusion models.
  • 🥇 Infosys Springboard Project: Developed PlantDocBot, a high-precision dual-tower agronomy diagnosis engine pairing ResNet34 (95.2% accuracy) with fine-tuned symptom NLP.
  • 🥇 IBM Hackathon: Architected an autonomous multi-agent clinical nutrition assistant orchestrating dietary biomarkers and medical reasoning constraints.
  • 🚀 Eylo: Engineered autonomous Chrome MV3 stealth outreach automation with non-fingerprintable humanized interaction heuristics and GPT-4o-mini lead qualification.

🤝 Connect & Collaborate

Gmail    GitHub    LinkedIn

"The best systems aren't just fast or accurate — they feel like magic to the user."

Profile Views

Pinned Loading

  1. Business-Entity-Resolution Business-Entity-Resolution Public

    🏢 High-precision merchant record linkage and entity deduplication pipeline combining TF-IDF n-gram blocking, Levenshtein distance, and gradient boosted rerankers.

    Python

  2. Eylo-Reddit-v2 Eylo-Reddit-v2 Public

    🤖 AI-driven Reddit outreach suite featuring a Manifest V3 Chrome extension with stealth DOM automation, GPT-4o-mini lead qualification, and shared Supabase deduplication.

    JavaScript

  3. resume-ranker resume-ranker Public

    📄 Enterprise RAG candidate ranking engine featuring dense FAISS search, sparse SPLADE-v3 expansion, two-part Cross-Encoder reranking, and Qwen 2.5 LLM reasoning.

    Jupyter Notebook 1

  4. Smart-Product-Pricing Smart-Product-Pricing Public

    🏷️ Enterprise multimodal pricing engine combining DistilBERT text towers, DINOv2 vision encoders, cross-attention fusion, and LightGBM with cloud Parquet datasets.

    Jupyter Notebook 1