ahmeedgamil AhmeedGamil

 ┌─────────────────────────────────────────────────────────┐
 │  I build mobile apps that ship, AI systems that think,  │
 │  and developer tools that save time.                    │
 └─────────────────────────────────────────────────────────┘

# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
#  PROFILE MODE
# ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

role:
  - AI Systems Engineer
  - Software Architect
  - Mobile Developer

base: Sana'a, Yemen

current_focus:
  - Production products across mobile, games, and developer tooling
  - AI infrastructure, RAG, memory, and MCP systems
  - Compiler-style generation, code intelligence, and structured workflows
  - Clean architecture, scalable UX, and failure-resistant system design

highlights:
  - Published Flutter package with 7,000+ downloads
  - Built mobile commerce apps, AI tooling, code retrieval systems
  - Shipped game work with Unity

ai_workflow_stack:
  primary:   Claude Code Opus 4.6
  secondary: Codex ChatGPT 5.3

motto: "Build useful systems. Keep them fast, clean, and reliable."

📊 At a Glance

🗺️ Journey

╔══════════════════════════════════════════════════════╗
║                       ROADMAP                       ║
╚══════════════════════════════════════════════════════╝

2020
│
▼
Began building video games
as a solo Unity developer
│
▼
2020–2023
│
▼
Built games independently using Unity,
C#, and game-focused systems thinking
│
▼
2021
│
▼
Graduated in Information Technology
Engineering
│
▼
2023
│
▼
Expanded into Flutter and AI engineering
│
▼
2023–Present
│
▼
Building production mobile apps, developer tools,
code intelligence systems, RAG pipelines, and AI
infrastructure

🧰 Core Stack

📱 Mobile

🤖 AI Engineering

🎮 Games & Native

⚙️ Backend

🛠️ Tools

🧠 AI Workflow

🚀 Selected Builds

Featured Start Order

1. liquid_glass_easy · 2. Hamsa · 3. Bolt AI · 4. Memory System for AI

Other listed projects are private repositories.

liquid_glass_easy

Number one cross-platform Flutter package for interactive liquid glass effects with 7,000+ downloads on pub.dev. → View on pub.dev

📱 Mobile Projects

Moka Sweets

Production e-commerce app where I patched critical security, maps, and performance issues for a large live user base. → View on Google Play

Hamsa

> Large-scale beauty and makeup shopping platform serving Saudi Arabia, Yemen, Oman, and UAE with isolated data, pricing, and currency. > Built guest cart merge, secure multi-step checkout, and market-specific payment flows. > Delivered glass morphism UI, hero animations, RTL Arabic support, real-time order tracking, and push notifications.

Alhamzi

> Premium fashion shopping app with polished browse-to-checkout flows. > Added search, filters, wishlist, secure checkout, real-time tracking, and push notifications. > Built smooth glass morphism UI and full RTL Arabic support.

Aura Pharmacy

Implemented shared cart feature enabling multiple users to collaborate on a single pharmacy order. → View on Google Play

Biqalati

Grocery and daily essentials e-commerce app with fast product browsing, cart management, and real-time order tracking. → View on Google Play

Alam Semsem Toys

Improved location storage and fixed currency switching through REST API optimization. → View on Google Play

Uturn Food Ordering

Integrated custom real-time push notifications for order status updates. → View on Google Play

🔌 RSA E-Commerce

Electronics commerce app for Arduino boards, microcontrollers, and maker hardware. Added dynamic language and currency switching, secure payments, and real-time backend sync.

🤖 AI & Developer Systems

⚡ Bolt AI — Flutter Clean Code Generator Using Compiler Architecture

╔══════════════════════════════════════════════════════════════╗
║  THE IDEA                                                    ║
║                                                              ║
║  Most AI code generation tools produce raw code directly     ║
║  from prompts. Bolt AI treats Flutter generation like a      ║
║  real compiler problem: blueprint in → validated arch out.   ║
╚══════════════════════════════════════════════════════════════╝

CORE PIPELINE
─────────────────────────────────────────────────────────────
  YAML Blueprint  →  Parse  →  Validate  →  Transform
                                                    ↓
                              Production-ready Flutter code
                                                    ↑
                                               Emit  ←

INTERNAL STRUCTURE
─────────────────────────────────────────────────────────────
  lib/compiler/
  ├── ir/              Intermediate Representation
  ├── parsers/         YAML + JSON blueprint parsing
  ├── validators/      semantic + architecture validation
  ├── transformers/    IR transformation layer
  ├── emitters/        19 code emitters
  ├── pipeline/        orchestration and factories
  └── compiler.dart    entry point

GENERATED FEATURE SHAPE
─────────────────────────────────────────────────────────────
  lib/features/product/
  ├── domain/
  │   ├── entities/
  │   ├── repositories/
  │   └── usecases/
  ├── data/
  │   ├── models/
  │   ├── datasources/
  │   └── repositories/
  ├── presentation/
  │   └── bloc/
  ├── di/
  ├── product.dart
  └── product_registry.dart

Built around an Intermediate Representation — reasons about structure before generating Dart files.
Semantic validators check types, duplicates, naming rules, reserved keywords, and required fields.
Architecture validators enforce clean architecture boundaries, repository rules, use-case design, and presentation-layer consistency.
Emitter layer generates entities, models, repositories, data sources, BLoC files, DI modules, registries, endpoints, and barrel files.
Supports single-feature compilation and batch generation across multiple blueprints.
Exposes compiler operations through MCP tools: compile_blueprint, validate_blueprint, list_blueprints, and preview_ir.
Turns Flutter clean architecture from repetitive manual setup into a reproducible compiler workflow.

🧠 Memory System for AI

╔══════════════════════════════════════════════════════════════╗
║  THE IDEA                                                    ║
║                                                              ║
║  Most AI assistants forget useful context between runs or    ║
║  rely on one flat memory store. This system models memory    ║
║  the way humans do: episodic, semantic, and procedural.      ║
╚══════════════════════════════════════════════════════════════╝

SYSTEM STRUCTURE
─────────────────────────────────────────────────────────────
  MCP Client  →  mcp_memory_server.py  →  MemoryManager
                                              ├── VectorEngine      episodic memory
                                              ├── StructuredStore   semantic memory
                                              └── FileStore         procedural memory

MEMORY HIERARCHY
─────────────────────────────────────────────────────────────
  Episodic Memory
  ├── storage: ChromaDB
  └── use: experiences, lessons, summaries

  Semantic Memory
  ├── storage: SQLite
  └── use: facts, preferences, project details

  Procedural Memory
  ├── storage: Markdown files
  └── use: patterns, skills, reusable workflows

RETRIEVAL MODEL
─────────────────────────────────────────────────────────────
  Memory  →  question embedding
          →  answer embedding

  Query  →  search both indexes
         →  merge matches
         →  return compact relevant memory

Dual-index Q&A search — a memory can be retrieved whether the query matches the question or answer side.
Episodic memory powered by vector search, semantic memory by structured facts, procedural memory by file-based patterns and skills.
Uses Sentence-Transformers all-mpnet-base-v2 embeddings and token-efficient retrieval.
Includes project tagging, similarity thresholds, fact indexing, and stateless tool design.
Exposes 20+ MCP tools for storing, recalling, deleting, and managing memories, skills, and patterns.
Compatible with Cursor, Claude Desktop, Kiro, Antigravity, and other MCP clients.
Makes long-running AI assistants more reliable through structured memory instead of raw context accumulation.

🔍 AST-Based Code RAG — Flutter / Laravel

╔══════════════════════════════════════════════════════════════╗
║  THE IDEA                                                    ║
║                                                              ║
║  Standard code RAG splits source files like text documents.  ║
║  This project treats code as structured logic with           ║
║  relationships, flow, metadata, and audience-aware retrieval.║
╚══════════════════════════════════════════════════════════════╝

INDEXING FLOW
─────────────────────────────────────────────────────────────
  Source Files  →  AST Parser  →  Structured Extraction
                                          ↓
                              Cross-file Analysis
                                          ↓
                              LLM Enrichment  →  3 Named Vectors  →  Qdrant

CHUNK MODEL                        PER-CHUNK REPRESENTATION
──────────────────────────         ──────────────────────────────────────
  One point per:                   Chunk
  ├── class method                 ├── description vector
  ├── standalone function          ├── developer-questions vector
  ├── class-level unit             ├── user-questions vector
  └── config / route block         ├── calls / called_by
                                   ├── routes / models / tables
  Never whole-file chunks          └── raw source payload
  Never mid-function splits

QUERY FLOW
─────────────────────────────────────────────────────────────
  User question  →  query expansion  →  vector + BM25 hybrid retrieval
                                                    ↓
                                          merge + rerank
                                                    ↓
                                      call-graph expansion
                                                    ↓
                                        context builder  →  final answer

Closes the semantic gap at index time, not query time.
Stores enriched semantic descriptions, developer questions, user questions, and graph relationships — raw source kept in payload for answer-time context.
Retrieval combines dense vectors with BM25 hybrid search, then expands through calls and called_by neighbors for flow awareness.
Each chunk carries rich payload metadata: parameters, return types, visibility, routes, models, tables, git blame, and commit history.
Qdrant used with named vectors and payload-aware search; incremental re-indexing keeps the index current.
Includes a RAGAS-style evaluation setup using LLM-as-judge scoring for faithfulness, relevancy, precision, and recall.
Designed for serious codebase understanding across frameworks such as Laravel and multi-project retrieval workflows.

🧬 DNA-Inspired Multi-Agent Architecture

╔══════════════════════════════════════════════════════════════╗
║  THE IDEA                                                    ║
║                                                              ║
║  Most multi-agent systems use optional pipelines:            ║
║  planner → coder → reviewer.                                 ║
║  This project asks: what if agent relationships were         ║
║  structurally enforced, the same way DNA base pairing        ║
║  is enforced in biology?                                     ║
╚══════════════════════════════════════════════════════════════╝

AGENT MAPPING
─────────────────────────────────────────────────────────────
  A = Architect       T = Tester
  G = Generator       C = Critic

TWO-STRAND MODEL
─────────────────────────────────────────────────────────────
  Generative strand:     A ──────────► G
                         │             │
  Verification strand:   T ◄────────── C

  A pairs with T  ·  G pairs with C  ·  Pairing is mandatory

SYSTEM FLOW
─────────────────────────────────────────────────────────────
  Raw input
      → A builds structured specification
      → T validates specification
      → G generates from verified spec
      → C critiques against that spec
      → Merge mechanism
      → Final output

SEQUENCE-AS-CONFIGURATION
─────────────────────────────────────────────────────────────
  ATGC  →  analyze-first        GCAT  →  generative-first
  TACG  →  constraint-driven    CGTA  →  adversarial-first

  Same agents · Different reasoning strategy

Maps DNA base-pairing rules into an AI collaboration model where complementary roles are enforced at the architecture level.
A-T acts as the fast structural pair for requirements and validation; G-C acts as the deeper generation and critique pair.
Key architectural rule: silence is invalid — critique is mandatory and generation is tied to a verified specification.
Creates a parallel dual-track reasoning model instead of a loose sequential pipeline.
The sequence itself becomes a programmable reasoning layer — same four agents, different cognitive strategies.
Includes experiment artifacts and scored runs comparing structured DNA-style reasoning against other approaches.
One of the clearest examples of thinking about AI systems: not just prompts and agents, but rules, structure, constraints, and failure-resistant collaboration.

📦 Published Work

Project	Description	Link
✨ liquid_glass_easy	#1 cross-platform Flutter package for interactive liquid glass effects · 7,000+ downloads	pub.dev
🎮 Keep Flip	Paid Steam puzzle game built with Unity and fluid-physics gameplay	Steam
🏆 Implant Mission	Audience Choice Award winner in a game jam	—

📬 Contact

Got a mobile app that needs to ship or an AI system that needs to think?

Let's talk — I'm always up for solving real problems with well-built systems.

📍 Sana'a, Yemen · 🌐 Arabic (Native) · English (Professional)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly