Skip to content
Available
Osama IftikharGenAI & LLM Engineer

LLM systems thatfinish the job.

GenAI & LLM engineer in Dubai — 4+ years shipping fine-tuned models, RAG and multi-agent systems to production for teams across the EU, UAE and US.

Watch my intro 2:12
GitHubLinkedIn
Production topology
01Verified
92%

task completion on live voice calls

500+ calls · Global Health Clinic
02Verified
~50%

faster average time-to-answer

Fine-tuned internal search · Hatzs Dimensions
03Verified
~3 in 4

routine inbound calls resolved without a human

Aerlou front desk
Enter work log

In my own words.

Who I am, the kind of AI systems I build, and what I care about once they are live — in about two minutes. Captions are on by default.

Built across models, retrieval, agents, and evals. Then shipped as one system.

Systems with a pulse.

These are running systems and CV-grounded case studies, not concept cards. Open any of them to inspect the work directly.

Underneath it all

Production discipline: observability, retries, evals, and rollback paths—the parts nobody demos.

Every system looks simple on the surface.

Built across teams, markets, and constraints.

A record of ownership—from fine-tuned model rollouts and multi-agent platforms to voice systems and data pipelines, for clients across the EU, UAE, and US.

Global Health Clinic

AI Engineer — Team Lead, Agent Development · Remote · Ireland / Spain / Portugal / Czech Republic

  • Built and deployed multilingual LLM agents (English, Spanish, Portuguese, Czech) handling patient intake, appointment scheduling and FAQ resolution across four country-specific clinic portals; led the agent development track end to end, coordinating localisation and rollout across all four markets.
  • Layered ElevenLabs speech synthesis and real-time background-noise suppression over the agents with mid-call tool calling — calendar booking, CRM writes, live data retrieval — so calls completed tasks rather than only answering questions; deployed over Twilio and validated across 500+ live calls at a 92% task-completion rate.
  • Built a RAG layer over clinical and administrative knowledge bases to keep responses grounded and suppress hallucination on medical-adjacent queries, with escalation to human staff on low-confidence intents.
  • Analysed patient interaction logs in Python (Pandas, NumPy) to locate conversation drop-off points and rework flow logic; patient engagement rose ~35% across regions after rollout.

Ryzon Labs

AI Engineering & Product Lead · Dubai, UAE

  • Own product and engineering direction across the studio's AI product portfolio — architecture, model selection, production deployment and client rollout.

Ryzon Labs — Aerlou

Product Lead · Dubai, UAE

  • Built core assistant logic: intent parsing from voice notes and text, reminder and task extraction, and structured writes to a persistent memory layer covering people, preferences and expenses.
  • Integrated Google Calendar, Meet and Gmail so the agent could schedule meetings and issue invites from a single chat message, plus an OCR and summarisation pipeline for receipts, bills and PDFs sent into the chat.
  • Built the call and SMS front-desk mode with intent-based call routing and CRM-synced appointment booking, resolving roughly 3 in 4 routine inbound calls without human handoff.
  • Sold and deployed the voice agent into production with multiple dental and healthcare clinics, real estate agencies and automotive dealerships.

Ryzon Labs — Aertonic

Co-Founder & CTO · Dubai, UAE

  • Co-founded and set technical direction for an AI operations assistant for construction: RFI drafting and auto-chasing, submittal review, compliance checking, document intake and classification, and change-order tracking — wired directly into Procore, Aconex and email rather than sitting as a separate tool.
  • Built compliance logic that flags non-compliant submittals before installation, with every outbound action gated behind human approval; early deployments cut documentation time by up to 65% and RFI cycle time by roughly 50%.

Ryzon Labs — Jordy

AI / Backend Engineer · Dubai, UAE

  • Built the core agent logic converting a natural-language ICP description into a coordinated multi-channel outbound campaign across LinkedIn, email and X, with conditional follow-up driven by reply sentiment and engagement signals.
  • Built the personalisation and evaluation loop — tone matching from historical outreach, automated pre-send research from LinkedIn, company sites and social profiles, and self-adjusting sequencing against reply and meetings-booked rates — cutting time from raw ICP to live campaign from days to under an hour.

SBRE UAE

AI Engineer · Remote

  • Built conversational AI agents for lead qualification using RAG over property listings and CRM records, with per-client language switching and tool calling to book viewings, answer property questions and flag hot leads without a human in the loop.
  • Built an SMS agent on the Twilio API using OpenAI function calling to check live inventory, trigger CRM workflows and send follow-up emails mid-conversation, with a RAG layer over product and policy documents keeping answers grounded rather than generated.
  • Wrote data extraction pipelines to pull and structure MLS feeds and market reports, cutting time from raw listings to usable analysis by around 40%.
  • Tracked resolution rate and conversation drop-off in Python and SQL to drive iterative prompt, retrieval and flow improvements.

Hatzs Dimensions

AI Engineer — AI Team Lead · Remote

  • Fine-tuned an open-source LLM for internal knowledge search, paired with a web-retrieval layer and an OCR step for scanned tables; built the dataset and preprocessing pipeline from raw internal documents, cutting average time-to-answer by roughly half.
  • Took the model to production on Azure ML, then ran several weeks of hyperparameter tuning and A/B testing across retrieval and prompting configurations against real production usage.
  • Built Power BI monitoring dashboards over Python exports tracking model performance, token consumption and adoption across internal teams.
  • Led development of DealerIQ, an AI automation platform for automotive dealerships — automated cold calling, personalised cold-email campaigns and lead scoring, cutting manual first-pass outreach effort by an estimated 70%; built multi-agent flows with OpenAI Assistants and n8n, integrated real-time vehicle valuations via the Manheim API, and shipped client-facing pipeline and conversion dashboards.
  • Set technical direction for the internal LLM knowledge-search programme and coordinated delivery from architecture through production rollout.

Fark Sky Technologies

Blockchain Engineer · Dubai, UAE

  • Founding team member on an NFT marketplace launch that recorded just over $1M in first-month sales, alongside a token launch that reached 10,000+ active users within its first week.

One engineer across the whole LLM stack.

The useful part is not knowing isolated tools. It is connecting models, data, applications, and infrastructure into one reliable, production-grade system.

MODEL26 tools

AI systems

Fine-tuning, RAG, and multi-agent orchestration.

GPTLlamaMistralLLaVABERTLoRA / PEFTFull fine-tuningRAG pipeline designPrompt engineeringFunction & tool callingMulti-agent orchestrationNLPDeep learningCNNOpenAI APIAzure OpenAI ServiceLangChainLlamaIndexHugging FaceElevenLabsWhisperHallucination-rate benchmarkingHuman-in-the-loop approval gatingModel monitoringGrounded-response designPatient & CRM data across jurisdictions
DATA18 tools

Data & ML

Frameworks, pipelines, and analysis.

PyTorchTensorFlowKerasHugging Face Transformersscikit-learnPythonPandasNumPyMatplotlibSQLJavaScriptNode.jsRPower BITableauStatistical analysisData pipelinesOCR & document processing
APP16 tools

Production engineering

Serving, versioning, and evaluation in production.

MLflowKubeflowDVCCI/CDDockerKubernetesAzure MLAzure FunctionsGitModel servingExperiment trackingModel versioningHyperparameter tuningA/B testingEvaluation (accuracy · hallucination rate · latency)Token-usage monitoring
SHIP07 tools

Cloud & infrastructure

Deployment, automation, and integration.

Microsoft AzureMySQLHadoopTwilion8nSalesforcePostman

Full inventory

Everything in the toolbox.

Scroll the column—each tool has shipped in something real.

GPT
Llama
Mistral
LLaVA
BERT
LoRA / PEFT
Full fine-tuning
RAG pipeline design
Prompt engineering
Function & tool calling
Multi-agent orchestration
NLP
Deep learning
CNN
OpenAI API
Azure OpenAI Service
LangChain
LlamaIndex
Hugging Face
ElevenLabs
Whisper
Hallucination-rate benchmarking
Human-in-the-loop approval gating
Model monitoring
Grounded-response design
Patient & CRM data across jurisdictions
PyTorch
TensorFlow
Keras
Hugging Face Transformers
scikit-learn
Python
Pandas
NumPy
Matplotlib
SQL
JavaScript
Node.js
R
Power BI
Tableau
Statistical analysis
Data pipelines
OCR & document processing
MLflow
Kubeflow
DVC
CI/CD
Docker
Kubernetes
Azure ML
Azure Functions
Git
Model serving
Experiment tracking
Model versioning
Hyperparameter tuning
A/B testing
Evaluation (accuracy · hallucination rate · latency)
Token-usage monitoring
Microsoft Azure
MySQL
Hadoop
Twilio
n8n
Salesforce
Postman

Ask my CV anything.

Answered live by Gemini from the 9 sections of my CV, with sources. It can still be wrong — the CV is the final word.

● ready· 0/12

Try one of these, or type your own question below.

Need someone who can carry AI from first sketch to production?

Fine-tuning, RAG, multi-agent and voice systems — shipped into production for regulated, real-world teams, and measured after release.

Start a conversationLinkedInGitHub