Hi, I'm Alif Elham.

BizScout · Austin, TX, USA (Remote)

• Led BizScout's transition into an AI-native M&A platform, embedding LLM search, conversational agents, and autonomous content generation that supported a fourfold increase in the Pro-tier price.
• Architected Scout AI, a 22-agent assistant with a five-stage semantic search pipeline across 50,000+ listings, reducing inference costs by approximately 17× through tiered model routing, real-time embedding synchronization, and Langfuse observability.

Optimizely · Dhaka, Bangladesh

• Deployed five AI agents and copilots, including OptiGPT, using custom LLMs, RAG pipelines, and generative AI models; the systems reduced support time by 40% and were used daily by 90% of employees.
• Reduced API costs by 30% through caching, prompt engineering, and optimized data pipelines, while delivering workshops on AI adoption, ML infrastructure, feature engineering, and compliance.

Invisible Technologies · San Francisco, CA, USA (Remote)

• Managed a team of 42 AI specialists and developed a structured testing framework that improved model performance by 18%.

Invisible Technologies · Remote, USA

• Designed RLHF datasets, refined reward signals, and conducted multi-turn instruction tuning, improving GPT-4 response accuracy by 15%.

Mahdy Research Academy · Dhaka, Bangladesh

• Trained 300+ researchers in deep learning, computer vision, quantum computing, and medical image processing, increasing research-project success rates by 30%.

The Odd Pro · Las Vegas, NV, USA (Remote)

• Doubled revenue by integrating intelligent automation into a SaaS platform, reducing manual processing time by 50% and improving operational efficiency by 40%.

Fiverr · Remote

• Delivered 40+ end-to-end enterprise solutions spanning machine learning, cloud, and full-stack engineering for clients across more than 20 countries.

CDPF · Dhaka, Bangladesh

• Led cross-functional teams in developing a nationwide data-driven training platform used by 10,000+ students, improving personalized learning effectiveness by 35% through recommendation systems.