Iqra Hameed
AI Engineer at Zylo | Computer Vision & LLM systems
Islamabad, Pakistan
Actions
Iqra Hameed is an AI Engineer at Zylo, where she builds production LLM systems, including hybrid-retrieval RAG, multi-provider agentic tool use and real-time voice agents. She is also a published computer vision researcher. Her work covers deep learning-based human behaviour recognition for sustainable smart cities, and ongoing research into improving YOLO-family detectors for small object detection in surveillance and remote-sensing imagery. She holds a BS in Computer Science from the University of Gujrat. There she led the Computing Society's Research & Programming team and mentored students in programming and research. Iqra is passionate about making AI systems reliable, testable and trustworthy enough for real-world use.
Links
Area of Expertise
Topics
What it takes to Trust an AI agent: Lessons from Building One that acts on your behalf
An AI assisstant that only answers questions is one thing. An agent that sends email, creates calender events and searches your private documents is another. When an agent can act, every design gap becomes a trust problem.
In this session I share lessons from building Aether, an open source agentic AI workspace. It answers from user's oen documents and acts in gmail, google calender and google drive by text and by voice also. Behind it sit 18 tools, an MCP client, a provider-agnostic LLM layer and a hybrid retrieval pipeline, deployed as three independent services.
This isn't a success story. It's what building it taught me about trust:
Enforce boundaries in the data layer, not the prompt. Documents are isolated per conversation in SQL. I'll also show a latent risk I found where the guard lived in Python instead.
Measure before you believe. On a labelled benchmark, hybrid search (vector + BM25 fusion) improved Hit@3 by 33%. The cross-encoder reranker everyone recommends made results worse. I published that result.
Silent failures are the dangerous ones. One bug made failed tool calls reach the model as successes. There was no crash and no log.
Actions need human gates. My agent could send email the moment the model decided to. Approval steps and audit logs belong in the design from day one.
Real-world access is regulated. Google's verification tiers for sensitive scopes show what production access to user data actually demands.
Avoid vendor lock-in. One interface across OpenAI, Anthropic, Gemini and Groq, with a mock provider so 159 tests run without API keys or data leaving the environment.
Attendees will leave with a practical checklist for building AI agents that organizations can trust with real data and real actions.
Privacy by Design for Computer Vision in Safety-Critical Spaces
Cameras plus AI can make public spaces safer, but they also process footage of real people, often children. In this session I share the design principles behind SwimAI, a pool occupancy monitoring system built on a custom-trained YOLO11 model:
processing on the edge, so video never leaves the site
storing counts and coordinates, not footage
no face recognition or identity linking
consent and retention rules for any training data
I'll also cover responsible positioning: why the system assists staff and never replaces a lifeguard, and why we deliberately don't claim "drowning detection." Finally, I'll cover why every number shown to users must come from the model rather than a mockup. These lessons apply to any team deploying vision AI where privacy and liability matter.
Iqra Hameed
AI Engineer at Zylo | Computer Vision & LLM systems
Islamabad, Pakistan
Links
Actions
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top