Muhammad Muaz

Summary

What I did

• Designed an explainable matching engine scoring careers 0 to 100 on weighted skill coverage, RIASEC interest correlation and Pakistani education level fit, with plain-language reasons and a Best Next Skill suggestion.
• Built multilingual skill normalisation with RapidFuzz fuzzy matching and English, Urdu and Roman Urdu aliases, so students can enter skills the way they naturally write them.
• Developed Ask Rukh, a RAG chatbot that answers from career content and the student's own results using BM25 retrieval, the Groq API with a local Ollama fallback, numbered citations and PII redaction before any cloud call.
• Built an ESCO occupation importer with O*NET interest enrichment (ISCO-08 to SOC crosswalk) and an evaluation module (Hit@3, NDCG@5, MRR) comparing the engine with four baselines and a LightGBM LambdaRank ranker.
• Delivered version 1.0 with all planned modules working and 113 automated pytest tests passing, with ruff and mypy checks clean; now testing, refining the interface and preparing a student pilot ahead of public launch.
• Tested the recommendation engine using test profiles with partial, real-world skill inputs to verify career matches and skill gap accuracy.
• Taking ownership of the full cycle from building the ML engine and testing to launching publicly for students across Pakistan.

Results

• Delivered version 1.0 with 113 automated pytest tests passing.
• Implemented PII redaction for cloud-based LLM calls.
• Developed a multilingual normalization system supporting English, Urdu, and Roman Urdu.
• Managed a three-person team for internal testing and development.