Building at the edge of applied AI.

Scalable ML platform with automated parameter prediction for EEG model benchmarking and intelligent ingestion workflows.

• Reduced testing time from 2–3 days to minutes
• Tested by 5+ researchers in production

Retrieval-Augmented Generation system for chest X-ray interpretation using CLIP cross-modal retrieval and Groq LLM.

• FAISS similarity search over MIMIC-CXR
• Structured radiology findings & impressions

Neuro-inspired graph-based temporal-spatial model combining GAT and LSTM for EEG cognitive state prediction.

• GAT + LSTM hybrid architecture
• Temporal-spatial EEG modeling

GNN-based optimization system for seizure detection on the TUH EEG dataset with hyperparameter optimization.

• 90% detection · 87% classification
• 84% early prediction on TUH EEG

Graph neural network–based predictive maintenance system for industrial crane systems using sensor data and digital twins.

• Sensor + digital-twin fusion
• Failure-mode early warning

Streamlit web app enabling OCR and document understanding from uploaded images via vision-enabled Groq LLM.

• Auto-structured Markdown output
• Document Q&A

AI-powered document Q&A system using LLMs and FAISS to answer questions from uploaded PDFs.

• FAISS-backed retrieval
• Context-grounded LLM answers

Spiking Neural Networks on the ISRUC-Sleep dataset, optimized for low-power edge computing.

• 82% accuracy
• 35% energy reduction

Voice/chat assistant using LLMs and multi-criteria decision algorithms for academic recommendations.

• Multi-criteria ranking
• Voice + chat interface

U-Net model for segmenting dental CT scans, including nerves, roots, and enamel structures.

• 3D volumetric preprocessing
• 3D Slicer / Dragonfly visualization

AI web app to explore Lahore's history using LLMs and FAISS semantic search over Wikipedia and PDFs.

• FAISS vector retrieval
• PDF upload + summarization

Attention-based decoder combining temporal, spectral and spatial features across 500+ subjects.

• 93.3% accuracy
• +2.3% over prior work

Full-stack secure app for audio diarization, transcription, feature extraction and chatbot querying.

• JWT auth + responsive UI
• Groq LLM querying

Fine-tuned Meditron LLM with LoRA/QLoRA to replicate USMLE-style clinical reasoning.

• 92%+ reasoning accuracy
• 40% less manual annotation

Data analysis dashboard for Olympic history with insights into medals, countries, athletes, and trends.

• Interactive medal & country insights
• Trend analysis across decades

Full-stack AI legal assistant for case research, document summarization and case law referencing.

• VectorDB semantic querying
• LangChain LLM workflows

EEG emotion recognition pipeline with preprocessing, artifact removal and frequency-band features.

• Interpretable outputs
• Streamlit dashboard

Streamlit + MySQL system for managing educational rubrics, CLOs, assessments, and student performance.

• End-to-end CLO tracking
• Structured assessment workflows

DCGAN to generate synthetic MRI scans for brain tumor analysis and data augmentation.

• >90% authenticity (radiologists)
• Improved generalization

AI chatbot for industrial crane operators, providing answers from operation manuals and troubleshooting guides.

• 75% faster task resolution
• Multi-language retrieval

WhatsApp chatbot to guide visitors on religious sites, prayer timings and local etiquette.

• Multilingual real-time messaging
• Webhook architecture

ANN classifying malignant vs benign tumors using structured tabular features.

• 99% on UCI dataset
• k-fold cross-validation