01Work
Things I've built.
A complete index of projects — from LLMs and AI agents to multimodal RAG and full-stack ML apps, with a from-scratch tensor library thrown in for the low-level fun. Most are open-source.
Showing 13 of 13
- 01
Tensorax
MLView project- Built a standalone tensor computation library from scratch with custom C++/CUDA kernels and Python bindings, without relying on NumPy, PyTorch, or cuBLAS.
- Implemented complete automatic differentiation (autograd) with computational graph and PyTorch-like API including neural network layers, optimizers, and loss functions.
- Developed and benchmarked optimized CUDA matrix multiplication kernels using tiling, shared memory, and memory coalescing techniques, measuring up to 2.3× speedup over NumPy on the matmul benchmarks included in the repository.
- Published on PyPI with a complete neural network training pipeline including SGD, Adam optimizers, and multiple activation functions.
- Available on GitHub with comprehensive documentation and examples.
- 02
TinyGPT
MLView project- Pretrained TinyGPT, a 95M-parameter LLM from scratch in PyTorch on a constrained RTX 3070 Ti (8GB VRAM), utilizing mixed-precision (AMP) and gradient accumulation to prevent OOM errors.
- Executed Supervised Fine-Tuning (SFT) on the Alpaca dataset to improve instruction-following capabilities.
- Deployed an interactive Streamlit UI on a free-tier CPU with dynamic model switching. Published the model on HuggingFace and shared on GitHub and Streamlit Cloud.
- 03View project
- Engineered a multi-agent orchestration platform using LangGraph to handle intent classification, planning, and automated tool execution.
- Built a scalable workflow engine with custom scheduling (cron/interval) and JWT authentication, deployed via Docker for enterprise-grade reliability.
- 04
Split.ai
MLView project- Architected a fault-tolerant receipt parsing engine with a multi-LLM fallback strategy (Gemini → Llama 3), so a failed parse retries on a second model rather than dropping the receipt.
- Implemented a graph-based debt simplification algorithm to minimize transaction volume and utilized Celery/RabbitMQ for asynchronous OCR processing.
- 05
QuillGPT
MLView project- Implemented a GPT decoder block with Self-Attention and Multi-Headed Attention using PyTorch.
- Developed and pre-trained two distinct models: Shakespearean GPT and Harpoon GPT. Deployed a Streamlit Playground for interactive model exploration.
- Leveraged FastAPI microservice for rapid development with Docker containerization for portability.
- Created Python scripts for training new GPT models and performing inference, so a new model can be trained and served without rewriting the pipeline each time.
- Designed and implemented a simple tokenizer for efficient text encoding and decoding. Wrote custom serialization and de-serialization methods for portability.
- 06View project
- Developed a Multimodal Retrieval-Augmented Generation (RAG) system integrating OpenAI’s CLIP and Whisper models for seamless cross-modal retrieval and semantic search across text, images, and audio, responding within 2-3 seconds.
- Deployed a Streamlit-based interface for searching, annotating and creating custom datasets and fine-tuning the CLIP model with configurable parameters for domain-specific applications.
- Implemented multimodal input support, allowing users to upload images, PDFs, and audio files or use real-time audio recording for enhanced interaction.
- Enabled URL-based image indexing and website content scraping for seamless integration and retrieval of web data.
- Utilized FAISS for efficient similarity search by aligning embeddings in a shared latent space.
- 07
ParkVision
WebView project- Developed an AI-driven parking analytics and customer segmentation dashboard utilizing the MVC architecture.
- Fine-tuned YOLOv8 with 0.968 mAP and EasyOCR with 85% accuracy for vehicle plate recognition, integrating the IDFY API for extracting vehicle details.
- Designed and created REST APIs using Django REST Framework and FastAPI.
- Designed and implemented Progressive Web App with downloadable capabilities, including frontend API caching.
- Leveraged PostgreSQL for data storage, Celery for background tasks, Redis for caching, and Docker for containerization.
- 08
DermaCare.ai
Web- Developed an AI-driven portal for early detection of skin diseases, targeting patients in rural areas and facilitating consultations with the nearest dermatologists.
- Created a comprehensive solution featuring both a web platform and a mobile application. Designed an intuitive user interface with Figma, implemented the web frontend using ReactJS, and built the mobile app with Flutter. The backend architecture was robustly developed using Django REST Framework.
- Trained a sophisticated computer vision model for skin disease detection utilizing TensorFlow, deployed via FastAPI, with image uploads managed through an S3 bucket.
- Enabled offline inference capabilities within the mobile application by integrating a Tflite model, ensuring accessibility in low-connectivity areas.
- Successfully executed the entire project within a 24-hour timeframe during a hackathon, demonstrating rapid development and deployment skills.
- 09
M2ConneX
WebView project- Orchestrated the complete architecture of M2ConneX, a comprehensive platform empowering MMCOE alumni to connect, network, and collaborate seamlessly.
- Designed and orchestrated the backend with the Django REST framework, fashioning an elegant frontend with React JS with Tailwind CSS and Docker for containerization, ensuring scalable and consistent deployment.
- Fortified the platform’s integrity with real-time checks, implementing image classification and a robust profanity checker with 95% accuracy to swiftly neutralize NSFW content in just 6-7 seconds.
- Engineered a sophisticated recommendation system using Scikit-learn to deliver personalized suggestions for connections, posts, and job opportunities, tailored to user skills and experience.
- Used PostgreSQL for data storage, Celery for background tasks, and Docker for containerization.
- 10
Sonnet
MLView project- Developed a chatbot that suggests songs based on user input by leveraging advanced language models and lyrics analysis.
- Implemented LangChain for efficient natural language processing and FAISS Vectorstore for fast, accurate similarity search in song lyrics, delivering rapid responses in 2-3 seconds.
- Integrated a database of songs and lyrics to enhance the recommendation engine's accuracy and diversity.
- Created and optimized Python scripts for processing text inputs and querying the song database.
- 11
DevFinder
App DevView project- Empowered college students and developers through an innovative social platform featuring AI-powered recommendations and real-time messaging.
- Spearheaded full-cycle mobile app development using Figma and Flutter, integrating advanced APIs and infinite scrolling for a buttery-smooth UX.
- Architected a powerful backend with Django REST Framework, paired with a polished ReactJS frontend for enterprise-grade scalability.
- Pioneered dynamic content loading and RESTful API optimizations, slashing latency by 65% during peak usage.
- 12View project
- Developed an email and text spam filter utilizing "ALBERT" an Encoder-Decoder model.
- Fine-tuned the ALBERT model on a spam detection dataset and deployed it on Hugging Face.
- Built and hosted a frontend interface using Streamlit for user interaction.
- 13
OpenNN
MLView project- Developed a comprehensive Neural Network Training library similar to TensorFlow 2 from scratch in Python.
- Implemented core neural network components including layers, activations, loss functions, and optimization algorithms.
- Install via PyPI:
pip install open-nn-python
- Created detailed documentation and example notebooks to demonstrate the usage and capabilities of the library.
