Faizan Ahmed Khan

• delivered production AI systems for national security in air-gapped environments. Architected complete ML lifecycle from
research to deployment, reducing project delivery time by 40%.
• Built offline English-Arabic neural machine translation system using M2M100 transformers fine-tuned on 500K+
sentence pairs. Improved BLEU score from 57% to 85% on government document translation benchmarks.
• Deployed translation system on air-gapped networks with zero external dependencies. Applied INT8 quantization
reducing model size from 2.4GB to 600MB (75% reduction) while achieving 3x faster CPU inference at 200ms
latency.
• Built internal online RAG system from scratch over 25K+ heterogeneous documents (PDFs, DOCX, JPG, PNG);
used Tesseract OCR for image-based text extraction and Ollama for local LLM inference with zero external API
dependencies.
• Established MLOps infrastructure with CI/CD pipelines.

• Designed hybrid recommendation system combining collaborative filtering and content-based approaches for Islamic
matrimony platform serving 100K+ daily users. Improved match precision by 35% and user engagement by 40%.
• Built conversational AI chatbot using Rasa NLU handling 50K+ conversations monthly with 87% intent accuracy.
Deployed on AWS Lambda with auto-scaling, reducing response time from 2s to 300ms.
• Developed facial verification system using ResNet50 with triplet loss achieving 94% accuracy on 50K+ verification
attempts. Reduced fake profile creation by 45%, improving platform trust score from 3.2 to 4.1/5.
• Established automated ML pipeline with Feast feature store, GitLab CI/CD, and TensorRT optimization handling
10K+ inference requests/min. Scaled platform from 50K to 100K+ DAU with 99.9% uptime.