Skip to main content
Backend Engineer - AI & LLM Integrations at Aligned Tech

Jad Akil

AI Engineer | Backend Developer

LLM Integrations & RAG Systems

I build retrieval-augmented learning tools, forecasting workflows, FastAPI services, and distributed systems.

View ProjectsContact MeDownload CV
Focus
  • LLM integrations
  • RAG pipelines
  • FastAPI & Node.js
  • Spring Boot & Kafka
  • PostgreSQL & Redis
  • Docker & AWS
AI & backend roles
3
Selected projects
6
Certified Cloud Practitioner
AWS

About me

Where AI meets backend engineering

I am an AI engineer and backend developer in Lebanon, with a computer science background from the Lebanese University. I currently contribute to Hader AI / Alignbot at Aligned Tech, a multi-tenant conversational AI platform for businesses across WhatsApp, Instagram, Messenger, and voice channels.

My experience spans course-aware learning tools, macroeconomic forecasting, LLM provider routing, secure webhook flows, and event-driven microservices. The projects below document the models, systems, and tests behind that work.

  • Grounded AI

    Retrieval, context filtering, and validation so model output stays scoped, explainable, and useful.

  • Backend systems

    Typed APIs, secure webhooks, multi-tenant access control, and queues that keep services reliable.

  • Measured results

    Tests, held-out evaluation, and documented workflows so every claim has evidence behind it.

Quick facts

Based in
Lebanon
Currently
Backend Engineer - AI & LLM Integrations at Aligned Tech
Education
Master 1 (M1), Computer Science, Lebanese University
Certified
AWS Certified Cloud Practitioner (CLF-C02)
Languages
Arabic (Native) · English (Advanced) · French (Intermediate)

Let’s talk about AI and backend engineering.

Experience

Where I have built and shipped

Three roles spanning conversational AI, macroeconomic forecasting, and course-aware learning.

  1. Backend Engineer - AI & LLM Integrations

    Current

    Aligned Tech

    Sep 2026 - Present

    Contributing to Hader AI / Alignbot, a multi-tenant conversational AI platform for businesses across WhatsApp, Instagram, Messenger, and voice channels.

    • Develop backend services using TypeScript, Node.js, Fastify, PostgreSQL, Prisma, Redis, and BullMQ.
    • Integrate OpenAI, Anthropic, and Groq LLMs for conversational AI, prompt workflows, provider routing, and response validation.
    • Build secure messaging and webhook flows, including HMAC verification, tenant resolution, message processing, and AI-generated responses.
    • Implement multi-tenant security and access control using JWT, PostgreSQL RLS, and role-based authorization.
    • TypeScript
    • Node.js
    • Fastify
    • PostgreSQL
    • Prisma
    • Redis
    • BullMQ
    • OpenAI
    • Anthropic
    • Groq

  2. AI Research Intern - Macroeconomic Forecasting

    FINAID Solutions Tech.

    Jul 2026 - Sep 2026Remote

    Developed and evaluated machine-learning and time-series models to forecast the monthly direction of global financial and macroeconomic indicators.

    • Engineered features, compared model behavior, and validated results across forecasting approaches.
    • Documented research workflows and findings in Python for reproducible experiments.
    • Python
    • Machine learning
    • Time series
    • Model evaluation

  3. AI Engineer

    LebNexis

    Feb 2026 - Apr 2026Remote

    Built an internal course-aware AI learning platform for approved university material, supporting concept explanation, guided problem-solving, and knowledge reinforcement.

    • Implemented embedding-based retrieval and context-filtered LLM inference to keep responses scoped to approved course content.
    • Developed and tested backend endpoints for queries, sessions, and response delivery.
    • Diagnosed quality issues through prompt and retrieval-filter refinements.
    • Embeddings
    • Retrieval
    • LLM inference

Skills & technologies

Tools for the whole system

From model evaluation to APIs, data, and infrastructure.

Programming

Languages for AI, backend, and systems work.

  • Python
  • Java
  • C#
  • C/C++
  • JavaScript
  • TypeScript

AI and Machine Learning

Applied ML, deep learning, NLP, and retrieval-augmented systems.

  • PyTorch
  • TensorFlow
  • scikit-learn
  • time series
  • NLP
  • RAG
  • LangChain
  • embeddings
  • Sentence Transformers
  • model evaluation
  • cross-validation

Backend and Systems

APIs, auth, event streaming, and distributed services.

  • FastAPI
  • Spring Boot
  • Node.js
  • Express
  • REST APIs
  • JWT
  • OAuth
  • Kafka
  • distributed systems

Data, Cloud, and Tools

Storage, caching, containers, and cloud fundamentals.

  • PostgreSQL
  • PostGIS
  • MongoDB
  • Oracle
  • Redis
  • Docker
  • AWS
  • Git

Selected projects

Built. Tested. Explained.

Six projects across AI, distributed systems, and geospatial engineering.

Explore GitHub

Deep Learning

Hybrid Neural-MCTS Chess Engine

AlphaZero-style engine with a policy-value ResNet and batched PUCT search

Live DemoGitHub

A hybrid AlphaZero-style engine that combines a policy-value ResNet, batched PUCT-MCTS, a 35% classical-evaluation blend, and tactical, repetition, progress, and chess-principle penalties.

  • Designed an 11,757,287-parameter network with 20 board-state planes, a 160-channel/24-block residual trunk, a 4,672-action policy head, and a separate bounded value head.
  • Implemented fast neural-policy ranking with legal-move and mate/threat checks, then complexity-triggered 30/64/96-simulation MCTS probes, confidence stopping, and safety fallbacks for hanging pieces and queen blunders.
  • Engineered dual-source learning from parallel self-play and Lichess PGN/Zstandard data: sparse policy targets, float16 NPZ shards, invalid-sample filtering, BLAKE2b deduplication, checkpoints, and arena/Elo promotion.
  • Built a 5M replay buffer with streaming capacity up to 75M samples, covered by 101 backend tests.
parameters
11,757,287
sample streaming capacity, up to
75M
backend tests
101
  • Python
  • PyTorch
  • MCTS
  • FastAPI
  • WebSocket
  • React
  • Vite
  • NumPy

LLM Application

AI Recruiter Assistant

End-to-end recruiting platform with explainable matching and bilingual AI interviews

Live DemoGitHub

An end-to-end recruiting platform with async FastAPI/SQLAlchemy, React, PostgreSQL/pgvector, Redis-backed bulk CV processing, and REST/WebSocket workflows for candidates, jobs, matches, interviews, and reports.

  • Engineered PDF/DOCX/TXT parsing with OCR fallback and bilingual rule/LLM extraction grounded against a 500+ skill catalog and the ESCO taxonomy, including negation, learning status, experience, and evidence tracking.
  • Created an explainable hybrid matcher weighting required, optional, semantic, experience, and seniority signals 55/20/15/5/5, capped by required-skill coverage and adjusted by bounded top-K LLM reranking.
  • Delivered Arabic/English text, voice, and video interviews with Ollama/OpenAI, faster-whisper, edge-tts, and WebRTC.
  • Hardened access with JWT rotation, RBAC/workspace isolation, rate limits, security headers, and audit logs; designed an Nginx topology with two API replicas, a CV worker, PostgreSQL/pgvector, Redis, and GPU-capable Ollama.
skill catalog
500+
matcher weights
55/20/15/5/5
API replicas behind Nginx
2
  • Python
  • FastAPI
  • SQLAlchemy
  • React
  • TypeScript
  • PostgreSQL
  • pgvector
  • Redis
  • Ollama
  • OpenAI
  • Docker

Distributed Systems

SmartOrder - Event-Driven Order and Delivery System

Nine Spring Boot services with an idempotent Kafka Saga

  • Architected 10 Maven modules: nine deployable services plus a shared event/DTO library, with Eureka discovery, Spring Cloud Gateway routing, JWT/BCrypt authentication, and role propagation.
  • Implemented an idempotent choreography Saga over four Kafka topics, including atomic stock reservation, inventory release, and refund compensation paths.
  • Applied database-per-service persistence with PostgreSQL, MongoDB, and Redis, and containerized Kafka/ZooKeeper and the service topology with Docker Compose.
  • 10 Maven modules
  • 4 Kafka topics
  • 41 JUnit tests
  • Java 17
  • Spring Boot 3.2
  • Spring Cloud
  • Kafka
  • PostgreSQL
  • MongoDB
  • Redis
  • Docker
GitHub

AI Agents

Multi-Agent MCP Bridge

A 21-tool Model Context Protocol bridge for safe multi-agent delegation

  • Built fail-closed Scope Contracts that reject read-only mutations and out-of-scope, forbidden, shared, or unsafe edits.
  • Isolated agent HOME/XDG/temp state and denied unapproved shell, network, plugins, and nested delegation.
  • Implemented SQLite WAL queues, locks, leases, heartbeats, crash recovery, retries, idempotency, overlap detection, and AES-256-GCM request encryption.
  • 21 MCP tools
  • AES-256-GCM request encryption
  • SHA-256 receipt-bound integration
  • Node.js 22
  • Model Context Protocol
  • Zod
  • SQLite WAL
  • Git worktrees
  • LangGraph
GitHub

Geospatial AI

CityMind Tripoli - Urban Intelligence Prototype

Arabic-first urban intelligence with geospatial layers and explainable scores

  • Built MapLibre/OpenStreetMap layers for neighborhoods, reports, assets, and routes, with Tripoli-bounded citizen reporting, GPS/manual-location fallback, safe photo metadata, and duplicate detection.
  • Added role-filtered municipality and emergency-agency queues.
  • Implemented City Pulse aggregation, explainable Green Score/Walkability, five weighted route modes, rule-based scenario/risk engines, and grounded unavailable-data behavior.
  • 5 weighted route modes
  • 56 API tests
  • SRID 4326 PostGIS models
  • Next.js 14
  • React
  • TypeScript
  • FastAPI
  • SQLAlchemy
  • Alembic
  • PostgreSQL
  • PostGIS
  • MapLibre
  • Docker
GitHub

Machine Learning

Coronary Heart Disease Risk Prediction

10-year CHD risk pipeline on the Framingham dataset

  • Removed preprocessing leakage by fitting clipping and normalization statistics on training data only.
  • Selected hyperparameters and pruning through GridSearchCV and 5-fold cross-validated balanced accuracy.
  • Reached about 69% balanced accuracy on the held-out set.
  • about 69% held-out balanced accuracy
  • 5-fold cross-validation
  • 70/30 stratified split
  • Python
  • pandas
  • scikit-learn
  • Decision Tree
  • GridSearchCV
GitHub

Education & credentials

A foundation to build on

2026

Master 1 (M1), Computer Science

Lebanese University

First year of the Master's program completed

2025

Bachelor's Degree, Computer Science

Lebanese University

Certification

AWS Certified Cloud Practitioner (CLF-C02)

Amazon Web Services

Verify on Credly

AI Cohort Training

Khwarizmi Company

Jun 2026 - Jul 2026

Cloud Training Cohort

DOT Academy

Nov 2025 - Jan 2026

Languages

  • Arabic Native
  • English Advanced
  • French Intermediate

Get in touch

Let’s build what’s next.

Let’s talk about AI and backend engineering.

EmailDownload CVGitHubLinkedIn
Email
jadakeel05@gmail.com
Phone
+961 71 78 71 24+961 81 53 11 54
Location
Lebanon
Website
jadakeel.dev