Mohammed Faizan

Led delivery of production AI systems and LLM workflows end-to-end, owning architecture-to-deployment. Improved reliability and sped up delivery by instituting structured design reviews and scalable engineering practices. Mentored engineers on RAG pipelines, LLM development, and prompt engineering to raise overall output quality. Coordinated with product and engineering stakeholders to ship AI features on tight timelines while reducing production issues and iteration cycles.

Built and improved RAG pipelines across OpenAI, Gemini, Llama2, and Mistral, increasing response accuracy by ~30%. Developed LLM-powered document intelligence workflows that reduced manual processing time by ~95%. Designed end-to-end AI systems covering data ingestion, retrieval, LLM processing, output validation, and evaluation. Created an eval harness to track accuracy, latency, and reliability across AI services. Deployed scalable AI services using Docker, Kubernetes, and GCP/AWS and automated data extraction/processing to improve operational efficiency by ~70%.

Built an AI SaaS that converts a generic resume into a job-specific, ATS-ready resume via a multi-stage LLM workflow (extraction → deterministic scoring → gap analysis → optimization). Made scoring explainable by pairing LLM extraction with deterministic TypeScript scoring across skills/keywords/experience/ATS match. Reduced hallucinated experience using an evidence-gathering questionnaire engine plus coverage checks. Hardened the pipeline with prompt-injection safety by isolating untrusted resume/JD content from system instructions. Created an LLM evaluation harness (schema validity, hallucination checks, scoring accuracy) and a feedback loop converting user edits into test cases. Instrumented AI observability (model, latency, cost, error tracing) and cut spend via input fingerprinting, cached regeneration, and regeneration caps. Shipped full SaaS solo: Next.js, MongoDB, NextAuth, Stripe billing (quotas, credits, refunds), admin dashboard, and Playwright PDF export.