DHIA_VISION v6.2.0 // model: moumeni-dhia.pt // device: cuda:0
fps 60 latency 14ms t+00:00 REC
live inference · construction feed 04 · algiers
x:0000 y:0000
detection · class: person · conf 0.998 · id:001

MOUMENIDHIA

$

I teach machines to see. 6+ years building real-time video analytics for construction sites — detection, tracking, re-identification across large camera networks — now wiring that perception into LLMs and agentic AI that act on what the cameras understand.

6+years · vision & ai
24/7models in production
10+product features shipped
edge→cloudjetson · deepstream · aws
scroll ▾ inference continues
class: identityconf 0.00bbox #about

$ whoami --verbose

// parsed from the subject's config file

# dhia.config.yaml
name: "Moumeni Dhia"
role: "Senior Computer Vision Engineer"
base: "Algiers, Algeria" # UTC+1, remote-friendly
focus:
  - computer_vision      # detection · tracking · re-id
  - ai_engineering       # training → deployment, end to end
  - agentic_ai           # llm agents over live video
  - prompt_engineering   # evals, tool use, orchestration
  - backend_and_devops   # fastapi · docker · ci/cd · mlops
currently: "Lead AI & CV Engineer @ Evercam Construction Cameras"
status: open_to_interesting_problems
humans_mentored: many
cameras_watched: a_small_army

The short version: I build the whole pipeline — from a raw RTSP stream on a muddy construction site to a model decision, an event, an API, and a dashboard someone actually uses.

The unfair advantage: most vision engineers stop at the model. I ship the annotation tooling, the retraining loop, the edge deployment on Jetson, the message queue, the backend, and the monitoring that keeps it honest at 3am.

The current obsession: giving language models eyes — agents that can search, summarize and reason over what a fleet of cameras has seen.

class: experienceconf 0.00track_id 002

$ git log --career --graph

// 3 commits on branch main — no reverts

a3f9c21 (HEAD → main, origin/lead)2019 — present

Lead AI & Computer Vision Engineer

Evercam Construction Cameras

  • Designed & deployed real-time models for detection, tracking, re-identification, PPE and behavioral analysis on live construction feeds
  • Built end-to-end ML pipelines: data collection → annotation → training → continuous retraining → deployment
  • Robust multi-camera tracking powering smart search and movement analysis across sites
  • Crane activity detection, material classification, pose estimation, luminance/outage monitoring, 360° human blurring
  • Optimized models for edge deployment on NVIDIA Jetson — real-time processing, lower latency
  • Scalable backend services & APIs: FastAPI, Pydantic, SQLModel, Postgres, TimescaleDB, PostGIS
  • MLOps workflows: model versioning, CI/CD, monitoring, production rollout at scale
  • Shipped major apps: Gate Report, Smart Search, Observations, PPE Analytics, Workflows, Site Agent
  • LLMs & AI agents in production — perception meets reasoning
  • Mentored junior engineers through reviews, architecture guidance and technical leadership
7d2e8b02019 — 2021

Computer Vision Engineer

Evercam Construction Cameras

  • R&D on deep learning models for detection, segmentation and tracking
  • Built data pipelines for dataset generation, annotation tools and benchmarking metrics
  • Developed the earliest generations of detection models and video-processing pipelines
1c0ffee2017 — 2019

Software Engineer & Mobile Developer

IT Synergy

  • Android apps for geolocation services in Java
  • Web applications with Laravel (PHP) and Vue.js, delivered across multiple industries
class: skill_graphconf 0.00nodes 9 · edges 8

$ render --pipeline production.dag

// the system I build and run — every node is production code, not a buzzword

CAMERASrtsp · 360° DECODEdeepstream DETECTpytorch TRACKmulti-object RE-IDcross-camera REASONllm agents EVENTSrabbitmq APIfastapi INSIGHTgrafana · apps // throughput: real-time · deployed on NVIDIA Jetson at the edge + cloud GPU // each packet you see moving = a frame becoming a decision

perception

PyTorchTensorFlowOpenCVobject detectionmulti-object trackingre-identificationsegmentationpose estimationfeature matchingNVIDIA DeepStreamJetson edge

genai & agents

LLMsagentic AIprompt engineeringtransformerstool-use orchestrationvision + languageevals

backend

PythonFastAPIPydanticSQLModelPostgresTimescaleDBPostGISRabbitMQRedis

devops & mlops

DockerCI/CDmodel versioningGrafanaLokiPrometheusS3 · EC2 · ECRSeaweedFS
class: appsconf 0.00instances 7

$ ls /production/apps --user-facing

// things real customers click on every day, powered by models I built

detection

Gate Report

Vehicle detection & classification at site gates — automated delivery logs from raw video.

re-id

Smart Search

Find a person or machine across an entire multi-camera site — re-identification at network scale.

monitoring

Observations

Automated visual observations across live feeds — events with camera evidence, tagged and timestamped.

safety

PPE Analytics

Helmet & vest compliance monitoring on live feeds — safety signals without a human watching.

privacy

360° Human Blur

Real-time human blurring in 360° cameras — privacy compliance baked into the stream.

custom vision

Workflows

You design what to spot on your site — zones, rules, conditions — and we ship it. PPE at high altitude with custom rules? Done.

agentic ai

Site Agent

LLM agents over site video — ask the cameras questions, get answers grounded in what they saw.

class: training_dataconf 0.00epochs 5+

$ cat education.log | head

// pre-training, before the fine-tuning that is production

M.Sc. Artificial Intelligence

USTHB · 2015 — 2017

Master's degree in Artificial Intelligence — where the vision obsession started.

B.Sc. Software Engineering

USTHB · 2012 — 2015

Bachelor's degree in Software Engineering — the foundation everything ships on.

Certifications: DeepLearning.AI Professional TensorFlow Developer · DeepLearning.AI Advanced TensorFlow Techniques · Deep Learning Specialization (Coursera)
class: contact_surfaceconf 0.00reachable: true

$ connect --to dhia

// the model is available for inference requests

dhia@vision:~
dhia@vision:~$ contact --list
→ email    dhiaislam.moumeni@gmail.com
→ work     dhia@curiowitty.com
→ linkedin linkedin.com/in/dhia-moumeni
→ phone    +213 779 756 709
→ base     Algiers, Algeria (UTC+1)
dhia@vision:~$
./send_email --now ./open_linkedin