UbaidUllah

Lead Generative AI Engineer

EncoderBytes Pvt Ltd • Fulltime

Oct 2025 - Present • 11 mos

6-month Generative AI Engineer internship under the Pakistan Software Export Board (PSEB), followed by promotion to a full-time, on-site Lead Generative AI Engineer role. Built the n8n Booking Email Automator — an end-to-end automation parsing unstructured booking emails into clean JSON via IMAP triggers and OpenAI extraction (addresses, vehicle types, datetime). Handled real-world edge cases, validated data before API dispatch, and rolled out in phases (test → semi-auto → full production) with Google Sheets logging for traceability. Developed Pharmapedia, an AI-powered pharmaceutical content prototype using FastAPI, TinyLlama, and Stable Diffusion to generate structured drug information and illustrations from OpenFDA data, with a scalable UI supporting multi-language batch content generation. As Lead Generative AI Engineer, responsible for architecting and deploying production LLM applications using RAG, LangChain/LangGraph, vector search (Pinecone/FAISS), prompt engineering, and MLOps CI/CD across OpenAI, AWS, Azure, and Anthropic cloud integrations.

Research Assistant (RF Sensing & Telecom)

The University of Glasgow • Freelance

Dec 2024 - Jul 2025 • 7 mos

Delivered 6+ GenAI and RF-sensing research projects as a Research Assistant, focused on multimodal retrieval-augmented generation and sensor fusion for 6G and radar applications. Key projects: RF-Sensing GPT (multimodal RAG combining vector+BM25 retrieval with CLIP spectrogram analysis, 93.23% radar accuracy, 13.1% faithfulness improvement); TelcoGPT (hybrid embedding framework for telecom Q&A, 0.89 MRR, 95% QA accuracy); RF-Spectrogram VLM (few-shot activity recognition using vision-language models, 84.2% activity classification accuracy); and Multi-Radar Attention Fusion (adaptive attention across 3 radar sensor types, 99.21% accuracy on 11 activity classes). Work resulted in a publication (TelcoGPT, University of Glasgow ePrints) and demonstrated applied research skills: retrieval system design, multimodal AI, domain adaptation, and rigorous evaluation methodology.

Data Science

Bytewise Limited • Fulltime

May 2024 - Oct 2024 • 5 mos

Completed a 5-month intensive Data Science internship covering ML, DL, and NLP, completing 90 hands-on tasks over 3 months progressing from basic to production-level complexity. Built a CNN-ANN model predicting antibiotic resistance in Salmonella Typhi, fusing 28,278 real clinical isolates (Lady Reading Hospital, Peshawar) with molecular descriptors — achieved 96.2% accuracy and AUC=0.975, outperforming a Decision Tree baseline (94.3%), with statistically significant age-dependent resistance patterns identified. Developed a Virtual Voice Assistant combining Whisper (speech-to-text), a 4-bit quantized Llava model for image description, and TTS streaming, with multilingual support and a Gradio interface — built as a real-time accessibility tool for visually and hearing-impaired users.