Sanaullah Yasir
Request CV

AI ENGINEER · PYTHON BACKEND DEVELOPER

AI systems.
Built to work.

I’m Sanaullah Yasir. I build retrieval systems, evaluation workflows, and Python backends—from the first experiment to cloud deployment.

Explore my work View GitHub Request CV
PythonFastAPIRAGAzurePostgreSQLDocker
FLAGSHIP SYSTEM

ORION

Deployed API
ARCHITECTURE WALKTHROUGH 01 / 04

Hybrid retrieval combines BM25 and TF-IDF to find relevant evidence.

Illustration · not live telemetry
+33.3%Retrieval recall
+15.2%Answer F1
4-stageResearch loop
Azure · PostgreSQL · CI/CD Open live API
01 /Measure the resultPublic + held-out evaluation
02 /Keep people in controlHuman-gated changes in ORION
03 /Ship the systemAPIs · containers · cloud

SELECTED SYSTEMS

From idea to system.

Explore the implementation, inspect the source, or try a deployment. Deployed projects and repository demos are labeled separately.

6 systems

01
Live API · Azure

FLAGSHIP · AUTONOMOUS AI ENGINEERING

ORION — AI Engineering Control Center

An automated RAG experimentation platform that creates retrieval hypotheses, runs controlled experiments, evaluates results, stores experiment history, and keeps a human approval step before changes are accepted.

9 + 4 public and held-out evaluation items Human-gated deployment decisions
InvestigateHypothesizeExperimentRememberApprove

Reported gains: +33.3% retrieval recall and +15.2% answer F1. Evaluation scope: 9 public and 4 held-out items; not a broad production benchmark.

  • FastAPI
  • SQLAlchemy
  • PostgreSQL
  • BM25
  • TF-IDF
  • Docker
  • Azure
  • JWT
View architecture
Problem

RAG changes are often accepted without controlled measurement or memory of failed ideas.

Approach

Investigate → hypothesize → experiment → remember → request human approval.

Verification

Retrieval recall and answer F1 across separate public and held-out evaluation sets.

Delivery

Docker images published through GitHub Actions and deployed to Azure Container Apps.

Open live API ↗ View repository ↗
02
Live application

DOCUMENT INTELLIGENCE

Enterprise RAG 2.0

A document intelligence application that processes PDF and TXT files and produces grounded answers using Gemini embeddings and ChromaDB retrieval.

  • Python
  • Streamlit
  • ChromaDB
  • Gemini API
  • RAG
View engineering details
Ingestion

Content-hash chunk IDs preserve unrelated documents and make re-uploads safe.

Safety

Sanitized uploads, file-size limits, relevance filtering, and API retry handling.

Open live app ↗ View repository ↗
03
GitHub project

SOFTWARE INTELLIGENCE

EngineerAI

A repository analysis platform that inspects source code and configuration, calculates a weighted readiness score, and generates prioritized development tasks.

  • FastAPI
  • Next.js
  • SQLAlchemy
  • Docker Compose
  • pytest
View engineering details
Analysis

GitHub repository or ZIP inspection with security, quality, testing, docs, and Docker checks.

Decision support

Weighted 0–100 scoring, task prioritization, and optional OpenAI, Gemini, or Anthropic summaries.

View repository ↗
04
Live application

APPLIED MACHINE LEARNING

AI Interview Coach

A Streamlit application that evaluates interview answers with TF-IDF and Random Forest, then returns clear STAR-based improvement guidance.

  • scikit-learn
  • TF-IDF
  • Random Forest
  • Streamlit
Open live app ↗ Repository ↗
05
GitHub project

RETRIEVAL INFRASTRUCTURE

Enterprise RAG Pipeline

A self-hosted retrieval API using PostgreSQL and pgvector, local embeddings, JSONB metadata, and HNSW indexing for semantic search.

  • pgvector
  • FastAPI
  • HNSW
  • SentenceTransformers
View repository ↗
06
Technical demo

AGENTIC WORKFLOWS

Autonomous Data Agent

A FastAPI surface for a LangChain ReAct agent designed to reason over data and call tools through an autonomous workflow.

  • LangChain
  • ReAct
  • FastAPI
  • OpenAI
View repository ↗

INDEPENDENT RESEARCH IN SMART-GRID AI

Evidence before claims.

An independent machine-learning study exploring smart-grid fault diagnosis and electrical load forecasting using 8,713 hourly observations and walk-forward validation.

Read the paper View DOI

EXPERIENCE & EDUCATION

A practical foundation.

Oct 2024 — Dec 2025Remote

AI/ML Intern

Minus 100 Degrees

  • Automated IT infrastructure monitoring with Python scripts.
  • Built SQL and Python pipelines for system-log analysis.
  • Developed a proof-of-concept RAG support chatbot.
  • Collaborated on predictive-maintenance model deployment.
CompletedAbbottabad

Higher National Diploma in Artificial Intelligence

University of Hertfordshire · affiliated with MIUC

Portrait of Sanaullah Yasir

ABOUT

Build. Test. Improve.

I’m an early-career AI engineer from Abbottabad, Pakistan, focused on building useful AI systems instead of isolated model demos. My work combines Python backends, retrieval, evaluation, APIs, databases, and cloud deployment.

I’m especially interested in RAG, AI agents, developer tools, and systems that keep humans involved in important decisions.

Elements of AI Anthropic AI Fluency 2 Kaggle certificates Sports & Adventure Club President

TECHNICAL TOOLKIT

Skills grouped by purpose.

Languages

Python · SQL

AI & retrieval

LLMs · Generative AI · RAG · Evaluation · BM25 · TF-IDF · ChromaDB · Gemini · OpenAI

Backend & data

FastAPI · REST APIs · SQLAlchemy · PostgreSQL · Streamlit

Cloud & engineering

Azure · Container Apps · Docker · Compose · Git · GitHub Actions · GHCR

Security

JWT · bcrypt · Input validation · Safe uploads

BASED IN ABBOTTABAD, PAKISTAN

Let’s build something useful.

For AI systems, Python backend projects, or engineering opportunities, get in touch.

Send email LinkedIn GitHub Request CV