Senior AI engineer · LLM systems · Agentic AI · RAG

I build AI that holds up after the demo.

I'm Asif Ejaz, a senior AI engineer in Lahore with 7+ years designing, building and running AI and ML systems across healthcare, cybersecurity and enterprise software. I've shipped 10+ agentic systems to production, an enterprise RAG platform used by 300+ employees, and a phishing classifier that screens millions of emails. I build client projects through CognitionSync, and I'm open to remote roles.

Discuss a project Discuss a role

Open to new projects and remote roles. Book a free 15 minute call →

A neural network, training in your browsertraining
Epoch
0
Accuracy
50%
Loss
0.693
Parameters
337
Loss per epoch
Fig. 1Two hidden layers of 16 units learn to separate two classes. Click the plot to add points of your own and watch the model adapt. Written from scratch, no libraries.

In numbers

99.9%

precision on live email traffic, phishing classifier

employer figure

1M+

security events a day through an agentic SOC platform

employer figure

70%

fewer HR queries after an enterprise RAG launch

employer figure

50

businesses onboarded to Axiom Square, which I built

live

7+ yrs

building AI and ML systems, since 2019

record

Every claim carries a chip: live opens the running product code or record opens the source employer figure cannot be linked

Built and still running

Products I designed, shipped and keep online

Each one is running today, so the chip takes you to the real thing instead of a screenshot.

Evidence

Measured on real projects

Three results I can show the working for: a model comparison, an audit of my own page, and a cloud migration. Pick a tab, or read how the release check works.

eval_notebook.ipynb real runs

The model with the best precision did not ship

Four embedding models tested on DocuMind's energy research corpus.

The release check is 10 real questions that must pass before any deploy. The higher precision models never retrieved one table value from a cheat sheet document, so the lower scoring model stayed in production.

View as table
ModelPrecision at 5Release check
Voyage 3 Lite (shipped)0.6910 of 10
Gemini Embedding 0010.787 of 10
Gemini Embedding 20.827 of 10
OpenAI Embedding 3 Small0.7789 of 10

What I can help with

AI engineering, integrations and FBR digital invoicing

AI work runs through CognitionSync. Invoicing, tax filing and activation run through Axiom Square, which I built.

Client work

Freelance and studio projects since 2023

Alongside full time employment. Client names are withheld unless the work is public.

Quoted word for word from public reviews on my Upwork profile. record on Upwork

Experience

7+ years building AI and ML systems

Most recent first. Figures reported by my employers are marked as such.

  1. May 2025 to now

    Senior AI Engineer

    VentureDive · Lahore

    • Enterprise RAG and intelligent document processing. An enterprise RAG platform on LangGraph, pgvector, Mistral OCR and Claude Sonnet across 5K+ proprietary documents for 300+ employees, with 95%+ retrieval fidelity and 70% fewer HR queries. employer figure
    • Agentic SOC automation (Clean Machines). One orchestrator with four specialist agents on AWS Bedrock Agents, Lambda and API Gateway, processing 1M+ security events a day and cutting investigation time by 60%. employer figure
    • Healthcare ML (Qualderm Partners). Appointment overbooking recommendations for a multi state dermatology network, an estimated $1M+ annual revenue opportunity, plus slot utilisation forecasts in Snowflake dashboards. employer figure
    • Cut LLM inference cost by 30% through model routing, prompt caching and context optimisation, and defined production evaluation criteria for retrieval fidelity, groundedness, latency and token spend. employer figure
  2. Feb 2024 to Apr 2025

    Senior AI/ML Engineer

    SlashNext, acquired by Varonis · US cybersecurity

    • BERT classifier for phishing detection in PDF attachments, holding 99.9% precision on live enterprise traffic while scanning millions of emails. employer figure
    • Training and evaluation workflows with explicit gates on precision, false positive rate and production reliability, and remediation of 10+ security vulnerabilities in Python dependencies.
  3. Jan 2023 to Aug 2023

    LLM Coding and Evaluation Contractor (OpenAI)

    Turing

    • Solved and reviewed advanced Python and algorithmic challenges used as training and benchmark data for OpenAI code generation models, and ranked LLM responses to produce preference data for fine tuning.
  4. Apr 2022 to Feb 2024

    Associate Machine Learning Engineer

    The Entertainer · Dubai

    • Natural language to SQL system on GPT 4 and LangChain that let non technical teams query enterprise schemas, cutting data request tickets by 80%. employer figure
    • Sentiment classification and clustering over 1K+ app reviews a month, routing feedback to business units, and mentoring 6 junior engineers on LLM application development.
  5. Jan 2019 to Apr 2022

    Junior Machine Learning Engineer

    Baltoro.ai

    • Model observability infrastructure tracking latency, usage and output quality, supporting 99.9% system reliability.
    • Computer vision. Trained and deployed YOLOv5 models for real time tennis player tracking, scoring and profiling.
    • Speech. Audio transcription pipelines on Hugging Face models with denoising and automated post processing.

Education

  • MS in Data ScienceInformation Technology University, Lahore · 2021
  • BS in Computer ScienceGIFT University, Gujranwala · 2019

Toolkit

  • Python
  • PyTorch
  • TensorFlow
  • Hugging Face
  • LangGraph
  • LangChain
  • LlamaIndex
  • pgvector
  • Qdrant
  • Docling
  • Mistral OCR
  • AWS Bedrock
  • SageMaker
  • Lambda
  • MLflow
  • Databricks
  • Apache Spark
  • Snowflake
  • Docker
  • scikit learn
  • YOLOv5

full history on LinkedIn

Writing

Research, articles and guides

What I have published, from a peer reviewed paper to practical guides. I also post weekly build logs on LinkedIn.

Questions

Questions people ask me

Who is Asif Ejaz?

Asif Ejaz is a senior AI engineer based in Lahore, Pakistan, with more than seven years designing, building and running AI and ML systems across healthcare, cybersecurity and enterprise software. He has worked at VentureDive, SlashNext (acquired by Varonis), Turing, The Entertainer and Baltoro.ai, and builds AI products through CognitionSync.

Can Asif Ejaz build AI agents and automation for my business?

Yes. He builds tool calling agents, document question answering and workflow automation that connect to existing systems such as CRMs, ERPs and databases, with logging, limits and evaluation so the result can be measured.

Who can integrate FBR digital invoicing with my software or Odoo?

Asif Ejaz built Axiom Square, digital invoicing software that sends invoices to FBR and PRA, and an open source Odoo 18 addon for FBR digital invoicing. Setup and activation are handled through Axiom Square at axiomsquarepk.com.

Does Asif Ejaz work with clients outside Pakistan?

Yes. Recent clients include a UK construction group, and he has worked for employers in the US and the UAE. His working hours overlap with UK mornings and US East Coast afternoons.

Is Asif Ejaz open to remote roles?

Yes. He is open to remote roles in AI engineering, ML engineering and applied data science with teams in the US, UK and EU.

Contact

Tell me what the system has to do.

Book a short call, or email and I reply within one working day, Pakistan time (UTC+5). Small paid pilots are a good way to start.

Book a 15 minute callFree, on Google Meet. 10 AM US Eastern, 3 PM UK
For a projectAgents, automation, document AI, integrations
For a roleRemote AI, ML or data science roles
asifejaz.me@gmail.com +92 302 6279504 LinkedIn GitHub Upwork CognitionSync Axiom Square