Haris Tahir Rana

Haris Tahir Rana

AI Engineer | Independent Problem Solver & Team Lead | LUMS CS

Based in: Lahore, PakistanMost recently: Teaching Assistant - Introduction to Arabic Language, Lahore University of Management Sciences (LUMS)Experience: 1 year

Hire me

High-impact AI Engineer with a track record of building complex systems, from 97% accurate medical diagnostics to multilingual voice-AI for accessibility. Expert in the 'Plan-and-Execute' agentic pattern and custom backend logic, Haris balances a preference for deep, autonomous technical work with proven experience leading cross-functional teams of 5+ to deliver scalable products.

Experience

Teaching Assistant - Introduction to Arabic Language

Lahore University of Management Sciences (LUMS)work2026 – May 2026

Summary

Served as the Teaching Assistant for the 'Introduction to Arabic Language' course, supporting student learning and managing administrative logistics.

What I did

  • Held dedicated office hours to provide direct academic assistance.
  • Managed daily attendance and participation tracking.
  • Coordinated exam logistics and invigilation.
  • Evaluated assignments and provided feedback on grammar and vocabulary.

Results

  • Managed core administrative operations and supported student learning for the student cohort.
  • Held dedicated office hours to provide direct academic assistance and clarify course concepts.
  • Coordinated midterm and final exam logistics, including invigilation.
  • Assessed and marked regular assignments, providing feedback on foundational vocabulary and grammar.

Machine Learning Engineer

CLAPP HOSPITALLahore, PakistanworkFeb 2026 – Apr 2026

Summary

Developed an automated diagnostic system to classify hypernasal vs. normal speech patterns, implementing a speaker diarization pipeline to isolate 5,349 pediatric speech segments from 352 minutes of raw clinic audio and eliminate adult assessor voice interference. Engineered a 45-dimensional acoustic feature extraction pipeline optimized to 30 key diagnostic markers; implemented patient-level cross-validation to prevent data leakage and secure a 91.16% baseline classification accuracy. Selected XGBoost over CNN-based deep learning architectures for interpretability and robustness on a constrained dataset, then applied majority-vote aggregation across patient segments to lift diagnostic accuracy by ~3%, achieving 96.97% patient-level accuracy and 98.08% sensitivity on a 99-patient held-out test set.

Results

  • Achieved 96.97% patient-level accuracy and 98.08% sensitivity on a 99-patient held-out test set.
  • Secured a 91.16% baseline classification accuracy.
  • Isolated 5,349 pediatric speech segments from 352 minutes of raw clinic audio.
XGBoostCNNSpeaker DiarizationAcoustic Feature Extraction

Full- Stack Web Developer Intern

ROZEE.PKLahore, PakistanworkJul 2024 – Aug 2024

Summary

Built a voice-enabled WhatsApp bot using Node.js and JavaScript that guided users through CV creation by asking questions via voice and processing multilingual voice responses. Used OpenAI and Azure Speech APIs for real-time voice-to-text and language translation, enabling inclusive access for users with limited literacy. Developed a separate chatbot-based CV builder using simple text messaging, integrated with via Laravel (PHP), to make resume creation intuitive and accessible for all users.

Results

  • Architected a complex voice-to-voice pipeline: Speech-to-Text (Azure/OpenAI) -> Translation -> LLM Processing -> Back-translation -> Text-to-Speech -> WhatsApp API integration.
  • Engineered the bot to handle end-to-end CV creation, successfully extracting and formatting all standard professional sections (Work History, Education, Skills) from unstructured voice conversations.
  • Enabled global accessibility by implementing a multilingual translation layer, allowing users to interact in their native language while the LLM processed data in English.
  • Optimized the end-to-end voice processing loop to deliver responses within 15-30 seconds, managing multiple API calls for transcription, translation, and synthesis.
  • Achieved an end-to-end voice processing latency of 15-30 seconds per message for a complex multi-step AI pipeline.
Node.jsJavaScriptOpenAIAzure Speech APIsLaravelPHPWhatsApp API

Skills

Technical

PythonPython
PytorchPytorch
Numpy
Pandas
Next.jsNext.js
GitGit
SQLSQL
LangChain
Scikit-Learn
TypeScriptTypeScript
ReactReact
Google AI Studio
MongoDBMongoDB
React NativeReact Native
Stata
Power BI

Languages

English
Urdu

Projects

POWER ZONE INVENTORY MANAGEMENT SYSTEM

Lead Developerprofessional
Designed and deployed an end-to-end enterprise inventory platform to replace a legacy spreadsheet system, digitizing workflows from factory receiving to order fulfillment and reducing inventory mismatches and tracking errors by over 90%. Engineered a secure transactional database engine using atomic PostgreSQL functions, row-level locking, and Row-Level Security (RLS) to guarantee data integrity and enforce strict role-based access control across 8 distinct modules.
ReactTypeScriptNext.jsPostgreSQL

HEALTHSYNC

Team Lead & Lead Developerpersonal
Led a 5-member team to design and build a HealthSync, a mobile app prototype connecting patients with verified doctors. Developed a custom backend scheduling algorithm to manage complex appointment logic, ensuring zero overlapping slots and balancing doctor availability with patient preferences. Engineered a secure provider onboarding workflow using Firebase Storage for credential verification and integrated an AI chatbot for support.
React NativeTypeScriptDjangoMongoDBFirebase Storage
View the code

TUBERCULOSIS DETECTION

Researcher / Developerpersonal
Designed and implemented a deep learning pipeline for automated TB detection, integrating UNet++ segmentation with SEResNet50 and Swin Transformer models trained on 5,000+ curated chest X-ray images across multiple public datasets. Achieved strong model performance with 99% classification accuracy (0.99 macro F1), a 96% Dice score (0.9173 IoU), and GradCAM++ explainability; authored a research manuscript documenting the methodology and findings.
PytorchGradCAM++UNet++SEResNet50Swin Transformer
View the code

AGENTIC 10-K INTELLIGENCE SYSTEM

Developerpersonal
Built an automated Q&A and document ingestion pipeline for SEC 10-K filings, implementing semantic partitioning and HTML-to-Markdown table parsing to maintain financial data integrity and boosting RAGAS Faithfulness from 0.18 to 0.74 using various prompting techniques. Engineered an autonomous Plan-and-Execute research agent to track trends across 50 multi-year filings, integrating a custom Python-based calculator tool for precise ratio analysis and a summarization engine that delivered investment memos with a 4.39/5.0 G-Eval coherence score.
LangChainChromaDBGoogle AI StudioPython
View the code

Education

BSc ( Hons.) in Computer Science

Lahore University of Management SciencesArtificial Intelligence SpecializationGPA 3.37Aug 2022 – Jul 2026

Coursework

Statistics

A-Level

Lahore Grammar SchoolAug 2020 – Jun 2022

Coursework

3 A* 1 BSAT Score - 1510

Recognition

Certifications

Power BI

Datacamp

Data Analytics Essentials

Cisco

AI Fundamentals

IBM

Patents & publications

Tuberculosis Detection Research Manuscript

research paper2025

Authored a research manuscript documenting the methodology and findings of a deep learning pipeline for automated TB detection.

Governance & advisory

Researcher

FINTRA-LUMSotherSep 2023 – May 2024

Authored and published "How Can the Ehsaas Program Be a Financial Revolution for the Masses of Pakistan?" in Bottomline, FINTRA-LUMS's annual finance magazine, analyzing the program's economic impact, financial inclusion mechanisms, and effectiveness in poverty alleviation.

Student Council Prefect

LAHORE GRAMMAR SCHOOLotherSep 2021 – May 2022

Organized and supervised major school events, including Society Fairs, Inter-School Competitions, Sports Fests, and Parent-Teacher conferences, engaging the entire 200+ student body.

What drives the work

Motivations

  • Building high-accuracy AI systems for real-world impact
  • Solving complex logic puzzles (like scheduling algorithms)
  • Empowering others through teaching and mentorship
Hire HarisReach out about a role, a contract or a conversation.For recruiters

Haris's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis