Hafsah Akram

I build practical AI systems that work in real-world settings, from LangGraph workflows and RAG pipelines to full backend integrations. My work is grounded in research, with a strong focus on turning insights into reliable, testable systems.

I handle the full cycle: data extraction, prompt routing, automated evaluation, and deployment. Passionate about building AI that's not just smart, but solid, maintainable, and built to last

AI Engineer at Acceleration Engg (2025-01 – Present)

Building practical AI systems and workflows

• Designed LangGraph prompt workflows for a specialized flight search agent, utilizing the ReAct (Reason + Act) framework to encode routing logic and prevent infinite execution loops, cutting LangSmith-traced errors by 86% and response latency by 33%
• Debugged malformed tool-message sequences causing OpenAI BadRequestErrors by enforcing strict JSON schema compliance
• Configured Redis rate-limiting and session ID to manage user interaction limits
• Built an RAG pipeline using Qdrant and hybrid BM25 semantic search to retrieve internal documents
• Iterated on system prompts to maintain strict content-filtering boundaries based on retrieval quality metrics
• Engineered a custom LLM-as-a-judge pipeline utilizing GPT-4o to benchmark the summarization accuracy of smaller models
• Designed deterministic evaluation prompts using Chain-of-Thought (CoT) reasoning and zero-temperature parameters to ensure repeatable grading

AI Intern at Yoan Systems (2024-06 – 2025-11)

Supporting AI and backend development

• Configured and tested a live chatbot using a WordPress plugin, writing the initial prompts to handle standard user queries
• Supported senior developers by writing basic SQL queries to assist with backend data validation

Master in Computer Engineering – University of Engineering & Technology (2021 – 2024)