Skip to content
@ Ahmed AfzalSenior AI Engineer at Sagaris

welcomewelcometotomymyrealmrealmofofshippedshippedAIAIsystemssystemsandandawkwardawkwardedgeedgecasescases

SAGARIS — product interface

SAGARIS

ORIVFY — product interface

ORIVFY

MIDL-DICOM — product interface

MIDL-DICOM

MEDSCRIBE — product interface

MEDSCRIBE

001 / AGENTIC SALES

SAGARIS

Senior AI Engineer · 2026 - Present

Visit Sagaris

AgenticAgenticmulti-touchmulti-touchrevenuerevenueOSOSthatthatautomatesautomatesoutboundoutboundsalessalesacrossacrossemail,email,SMS,SMS,LinkedIn,LinkedIn,andandsocialsocialchannels.channels.

  • Python
  • LLMs
  • Agentic AI
  • MCP
  • FastAPI
  • n8n
  • Postgres

What I built

  • Engineered the multi-touch campaign orchestration engine driving 15+ step cross-channel sequences with per-contact channel deferral and reply-driven re-routing.
  • Built the AI hyper-personalization engine generating context-aware outreach per prospect.
  • Developed an in-house domain and mailbox warming subsystem for large-scale sending.
  • Delivered agentic LLM features for the intelligent dialer with intent scoring and a real-time AI sales coach, orchestrated via MCP-based tooling.

Hard parts

  • Coordinating stateful, long-running sequences across channels without double-touching a prospect.
  • Keeping deliverability high while scaling send volume across many domains and mailboxes.

Outcome

  • Lifted reply rates by 30% through cross-channel orchestration with reply-driven re-routing.
  • Improved positive-reply rates by 25% over templated baselines via hyper-personalization.
  • Sustained 98%+ inbox deliverability across large-scale sending.

002 / AI DETECTION

ORIVFY

Lead AI Engineer · 2026 - Present

Visit Orivfy

Real-timeReal-timedetectordetectorforforAI-generatedAI-generatedcontent,content,analyzinganalyzingthetheuniqueuniquefingerprintsfingerprintsleftleftbehindbehindbybylanguagelanguagemodels,models,imageimagegenerators,generators,andandvideo-synthesisvideo-synthesistools.tools.

  • Python
  • PyTorch
  • Transformers
  • Computer Vision
  • Explainable AI
  • FastAPI

What I built

  • Built multi-modal detection pipelines for text, image, and video content with granular classification.
  • Implemented passage-level and sentence-level AI classification for higher precision text analysis.
  • Integrated provenance tracking and explainability layers into the core product architecture.

Hard parts

  • Balancing detection accuracy with false-positive rates across diverse generative model families.
  • Designing interpretable outputs for enterprise users requiring audit-ready AI safety tooling.

Outcome

  • Improved text-detection accuracy by 18% through passage-level and sentence-level classification models.
  • Reduced false positives by 22% via explainable-AI and provenance layers built into the product architecture.
  • Served detection at sub-300 ms p95 latency across multi-modal fingerprint analysis.

003 / MEDICAL IMAGING

MIDL-DICOM

Associate AI Engineer at NCAI · 2024 - 2025

Visit MIDL

IntelligentIntelligent2D2Dandand3D3DDICOMDICOMviewerviewerwithwithvoice-basedvoice-basedinteractioninteractionandandVLM-poweredVLM-powereddiagnosticsdiagnosticsforforradiologyradiologyimaging,imaging,builtbuiltforforNCAI'sNCAI'sMedicalMedicalImagingImagingandandDiagnosticsDiagnosticsLabLab(MIDL).(MIDL).

  • Python
  • DICOM
  • Cornerstone
  • Computer Vision
  • VLMs
  • Voice AI
  • FastAPI
  • React

What I built

  • Developed intelligent 2D/3D DICOM visualization with multimodal AI analysis capabilities.
  • Integrated voice-based interaction for hands-free radiology workflow assistance.
  • Connected vision-language models for diagnostic support and imaging interpretation.

Hard parts

  • Handling large medical imaging datasets with strict latency and compliance requirements.
  • Building trustworthy AI assistance suitable for clinical decision support environments.

Outcome

  • Agentic imaging platform supporting radiology workflows with multimodal AI interaction.
  • Deployed research-to-production pipeline for medical imaging AI at NCAI.

004 / HEALTHCARE AI

MEDSCRIBE

AI Engineer II at NCAI · 2025 - 2026

Visit MedScribe

MedicalMedicaltranscriptiontranscriptionandandreport-automationreport-automationtooltoolintegratingintegratingASRASRandandNLPNLPwithwithclinicalclinicalsystems.systems.

  • Python
  • ASR
  • NLP
  • FastAPI
  • EHR
  • PACS
  • Docker
  • Kubernetes

What I built

  • Built real-time speech-to-text pipeline optimized for clinical vocabulary and ambient documentation.
  • Automated structured medical report generation using NLP post-processing and templating.
  • Integrated with hospital EHR and PACS systems for end-to-end clinical workflow embedding.

Hard parts

  • Achieving reliable transcription accuracy in noisy clinical environments with domain-specific terminology.
  • Meeting healthcare compliance, security, and deployment constraints for production hospital use.

Outcome

  • Raised transcription accuracy to 94% and cut report turnaround time by 65%.
  • Sustained 99.9% uptime across live production clinical environments.
  • Drove adoption across 3 hospital departments through direct clinician validation.

That's all for now.

GotGotaaprojectprojectininmind?mind?Let'sLet'stalktalk

Get in touch
Email:
ahmadafzalch007@gmail.com
Phone
+92 308 9469428
Based in
Islamabad, Pakistan