Building AI for autonomous systems — and exploring how machines should communicate confidence, uncertainty, and reasoning when human judgment is on the line.
I'm an AI engineer moving from machine learning toward embodied intelligence — designing systems that don't just generate answers, but help people decide when to trust them.
After two years building production LLM systems, RAG pipelines, and multi-agent architectures at Mutare Group, I'm now a Research Associate in Blue Marble Space's Young Scientist Program — designing future human habitats for the Moon and Mars that could support ~100-person communities, guided by planetary architecture, in-situ resource utilization, and real engineering constraints, and brought to life through AI-assisted visualization for education and outreach.
Woven through it all is one research question I keep returning to: how should AI communicate confidence and uncertainty so that humans — founders, habitat crews, anyone — know when to rely on it and when to question it.
Alongside research, I'm hands-on in the lab: simulating autonomous robots in Webots and ROS 2, prototyping embedded systems on ESP32 and Jetson Nano, and sharing everything I learn in public along the way — the wins, the failures, and the small victories in between.
One thread runs through all of my work: how should intelligent systems behave when humans must rely on them under uncertainty?
How AI systems should communicate confidence, uncertainty, assumptions, and risk — so people develop an accurate sense of when to rely on them and when to question them. Grounded in explainable AI and trust-calibration literature.
Learning the embodied side of intelligent systems step by step — starting with finite-state-machine controllers and sensor-driven navigation in Webots, and slowly working toward real hardware with ROS 2.
Decision support for habitats and field systems operating under communication delay — where operators can't phone home, and calibrated machine confidence becomes a life-support question.
Every module is real, built, and documented — from production LLM systems to robot controllers and embedded habitat experiments.
An early-stage independent research project asking how AI should communicate confidence and uncertainty, not just answers. Right now I'm in the paper-study phase — reading work on calibrated verbalized confidence (Mündler et al.) and trust calibration — and sketching the first ideas. No results yet; I'm sharing the journey as it unfolds.
My Blue Marble Space YSP research: designing an original habitat concept for a ~100-person community — surface and subsurface structures shaped by planetary architecture, ISRU, and real engineering constraints — visualized with design tools and AI imagery for education and outreach. Alongside it, I'm researching confidence-aware decision support for habitat crews; a co-authored abstract is submitted to Space Habitat Event 2026.
A seven-state finite state machine driving an e-puck through obstacle navigation, 180° turns, and IR-sensor wall following — with hand-tuned sensor thresholds and turn timing. Built for CU Boulder's robotics specialization; full lab report and simulation video recorded.
Ground-sensor line following fused with wheel-encoder odometry to estimate pose while tracking a curved path — the second Webots lab in the CU Boulder specialization, extending FSM work toward continuous sensor-driven control.
Practicing hands-on hardware building with ESP32 while assembling a ROS 2 setup across my laptop, Jetson Nano, and Arduino Mega — my step-by-step path from simulation toward real robots. Early days, and I'm documenting every step.
An AI-powered agent that automatically classifies, prioritizes, and routes incoming support tickets — smart classification across categories like authentication, billing, and technical issues, with priority prediction and auto-tagging for faster response times. Built with Python, scikit-learn, and pandas.
An offline-first LLM system built on Ollama with context-aware prompt chains, session memory, and a live debugging UI — replacing API calls with local execution to cut inference cost and latency.
Hackathon team project: an OpenAI-powered platform that parses resumes against job descriptions and returns SWOT analysis, skill gaps, and role-alignment recommendations through optimized few-shot prompt pipelines.
A real-time computer vision system that catches me slacking — custom-labeled dataset, fine-tuned YOLO detector, and an alerting layer that responds to prolonged distraction. That screenshot is it working live, mid-coding-session.
Final-year project: TensorFlow models predicting educational outcomes from socio-economic data across 639,000+ student records — surfacing five key indicators, served through FastAPI and an interactive Streamlit dashboard.
I'm learning robotics from the ground up, simulation-first — and documenting every step honestly. This is the actual state machine from my first Webots e-puck controller: seven states, tuned through trial, error, and a lot of patience.
Designing future human habitats for the Moon and Mars supporting ~100-person communities — analyzing planetary architecture across research, concept art, and design competitions, and developing an original surface + subsurface habitat design guided by ISRU and life-support constraints, communicated through AI-assisted visualization for education and outreach. Alongside, early-stage research into confidence-aware decision support for habitat crews, with a co-authored abstract submitted to Space Habitat Event 2026.
Designed and deployed production RAG systems with LangChain + FAISS/ChromaDB, multi-agent pipelines for research and decision support, local inference with Ollama, and end-to-end LLM evaluation and monitoring — from data ingestion through retrieval, inference, and deployment.
Built Generative AI features on LLM APIs for automation and NLP applications; integrated LLM components into full-stack systems and contributed to ML + LLM architecture design.
Selected to represent Pakistan (Bahawalpur), organizing regional events to bring the WiDS Worldwide conference to the local community. Also Global Ambassador at Women Tech Network and Women-in-Tech Associate at GDSC, The Islamia University of Bahawalpur.
Machine Learning, Deep Learning, NLP, Data Visualization, Data Structures & Algorithms. Final-year project: educational outcome prediction across 639,000+ student records.
"A Confidence-Aware Decision Support Framework for Space Habitat Operations Under Communication Delay" — co-authored with my research mentor, submitted.
AI solution leveraging GPU acceleration for real-time inference.
Top rank in exploratory data analysis, visualization, and statistical storytelling.
Documenting my journey from AI engineering into robotics and research: project breakdowns, papers explained simply, and honest lessons from failures — on LinkedIn, Instagram, YouTube, and Medium.
I'm interested in building intelligent autonomous systems capable of operating in complex, remote, and communication-constrained environments — embodied AI, robotic autonomy, and technologies that enable exploration beyond conventional terrestrial settings.
Open to research collaboration, AI engineering roles, and conversations about trustworthy autonomy.