DocIntel — Document Intelligence API
End-to-end document intelligence pipeline. Upload PDFs or images → layout detection, OCR, entity extraction → structured JSON. FastAPI + React + 65 tests.
ML systems, data infrastructure, and applied AI. 19 open-source repositories with real benchmarks and production code. GitHub has everything.
DocIntel — Document Intelligence API
End-to-end document intelligence pipeline. Upload PDFs or images → layout detection, OCR, entity extraction → structured JSON. FastAPI + React + 65 tests.
RenAIssance — GSoC 2025
End-to-end Document AI for Renaissance manuscripts. LayoutLMv3 layout detection, U-Net text recognition, Tesseract OCR, and post-correction. 179 files.
Vesuvius Challenge
3D deep learning for detecting ancient papyrus in X-ray CT scans. Four architectures (3D U-Net, Attention U-Net, V-Net, Double U-Net) with ensemble + TTA.
AgriFinance
Full-stack agricultural finance platform. ML credit scoring, satellite NDVI crop monitoring, climate-adjusted loans, and offline-first PWA for low-connectivity regions.
Swarm Robotics Simulator
Full-stack web app for multi-robot simulation. React + Express + PostgreSQL with live visualization, configurable presets, and performance analytics.
Flint Programming Language
Custom language with three backends: tree-walk interpreter, C transpiler, and LLVM compiler. Lexer, parser, AST, and code generation from scratch.
CampusBuddy
Flutter mobile app for campus life — events, dining, resources. 412 files across iOS, Android, and desktop with real data integrations.
ReAct agent with 4 memory modes. Chunked retrieval recovers needle-in-haystack from 0% to 100%.
Document AI for Renaissance manuscripts. LayoutLMv3 detection, Tesseract OCR, and post-correction pipeline.
Leakage-safe evaluation of Compositional Function Networks for biomedical tabular prediction. AUROC/AUPRC benchmarks.
CNN-RNN art classification. Style, artist, and genre prediction with confusion matrix validation.
Behavioral analysis for suicide prevention and mental health crisis detection with geospatial trend mapping.
Deep learning for detecting ancient papyrus surfaces from 3D X-ray CT scans.
End-to-end document intelligence: layout detection, OCR, and entity extraction via FastAPI. React dashboard, 65 tests, Docker deployment.
3-node distributed object store with erasure coding, LakeFS versioning, and Prometheus + Grafana monitoring.
Multi-domain visualization: volumetric scalar fields, CFD flow simulation, and 3D LiDAR point clouds.
Healthcare IoT pipeline connecting glucose monitors and sensors to BigQuery with Vertex AI analytics.
Multi-sensor robot training data validation webapp with embedded Rerun.io visualization.
Bimanual teleoperation platform for MyArm M&C arms with GUI visualization and real-time streaming.
Real-time robot vision with Modal-hosted SmolVLA inference and local camera input.
Browser-based multi-robot sim with flocking, foraging, and formation presets plus analytics dashboard.
RL motion imitation for Unitree G1 humanoid dynamic locomotion in MuJoCo.
Financial services for smallholder farmers. Climate-integrated credit scoring and risk assessment.
Infrastructure health monitoring platform using robot-collected sensor data for decision support.
ERD, ORM, and GUI system for a Doctor Who knowledge base. Database systems final project.
Custom interpreted language with interpreter, C transpiler, and LLVM compiler backend.
A deep dive into the four critical bottlenecks slowing the robotics revolution and why general-purpose robots remain inevitable
A deep dive into how Partial Information Decomposition (PID) reveals how different modalities interact in AI systems, from redundancy to synergy