Alizhan B.

Key Skills

Software

No software listed

Top Subject Matter

LLM-based career guidance and RAG reliability

Medical imaging diagnosis (leukemia screening) and federated model improvement

Medical imaging computer vision (classification + segmentation) reliability

Top Data Types

Text

Image

Document

Top Task Types

Question Answering

Diagnosis

Segmentation

Freelancer Overview

AI Architect & Co-Founder — SanaPath AI. Brings 1+ years of professional experience across legal operations, contract review, compliance, and structured analysis. Core strengths include FastAPI, Azure, and OpenAI SDK. Education includes Bachelor of Engineering, Heriot-Watt University / Zhubanov University (2028). AI-training focus includes data types such as Text, Medical, and DICOM and labeling workflows including Question Answering, Diagnosis, and Segmentation.

Labeling Experience

CTO & Co-Founder — AI Architect at Umit AI

Diagnosis

Architected an AI diagnostic pipeline for leukemia screening using an EfficientNetB3 image classifier combined with MedGemma 4B. Treated false-negative risk as the primary constraint and implemented clinician-facing interpretability via Grad-CAM. Built a federated learning framework to improve models across multiple clinics without transferring patient data. • Produced model outputs for clinical diagnostic decisions (screening) • Enabled explainability artifacts (Grad-CAM) for clinician audit of attention • Configured federated learning workflows for multi-site model improvement • Integrated offline edge inference with HL7/FHIR clinical systems

2026 - Present

AI Architect & Co-Founder — SanaPath AI

TextQuestion Answering

Designed and evaluated a dual-LLM pipeline (GPT-4o + Claude 3.5 Sonnet) to generate career guidance responses for 60,000+ students. Built a RAG system ingesting GitHub/LinkedIn data and added explicit handling to reduce noisy or adversarial input effects. Tuned multi-turn prompt logic to improve reliability and reduce hallucinated or overconfident advice in a high-stakes domain. • Generated and optimized prompt/response behaviors for student queries • Implemented structured benchmarking to assess factual reliability and consistency • Added guardrails for noisy/adversarial user input • Reduced hallucinations via iterative prompt and conversation design

2026 - Present

AI/ML Engineer, Competition Team — AI in Healthcare Hackathon 2026

Segmentation

Built EfficientNet-B5 classification and U-Net segmentation models for medical imaging in a healthcare hackathon competition. Applied Test-Time Augmentation (TTA) and Grad-CAM to help surface model uncertainty and areas of focus. Delivered results under competition evaluation while discussing reliability trade-offs with clinical-safety jurors. • Developed training workflows for medical image classification and segmentation • Used TTA to improve robustness at inference time • Generated Grad-CAM visualizations to support reliability and uncertainty assessment • Participated in clinical-safety discussions on failure modes

2026 - 2026

Education

Work History