AI/ML Engineer & Generative AI Developer — bridging deep technical expertise with business strategy to build robust RAG systems, fine-tuned LLMs, and end-to-end ML pipelines that create real, measurable impact.
AI workplace assistant that translates quantitative HR research into real-time psychological guidance. Processes 4,794 documents through a hierarchical chunking pipeline (2048→512→128 tokens) with hybrid dense+BM25 retrieval and ensemble reranking using Cross-encoder + Llama-3.3-70B. Built with enterprise safeguards: PII detection, toxicity filtering, prompt injection prevention.
Parameter-efficient Small Language Model that generates actionable marketing insights with two architectural approaches: an Encoder-Decoder (Tiny-T5) fine-tuned on instructional data, and an Encoder-only model with 3-phase training — domain adaptation → task fine-tuning → DPO alignment. CPU-runnable, low-cost inference.
Binary classifier enabling proactive HR intervention before employee resignation. Solved an extreme 87:13 class imbalance using SMOTE-ENN hybrid resampling, applied PCA retaining 95% variance across 30+ features, and tuned decision threshold to 0.35 for maximum recall — correctly flagging 24 of 39 at-risk employees.
Led a team research study examining how toxic positivity suppresses workplace creativity and problem-solving in Lahore's banking sector (n=135, convenience sampling). Key finding: psychological safety acts as a significant moderator of the negative effects of toxic positivity — providing actionable guidance for HR leaders and organizational strategists.
Open to AI/ML engineering roles, research collaborations, and internship opportunities.
Based in Lahore, Pakistan — available remotely worldwide.