stuff i've built
39 projects
LLM & RAG · 2026
Phi-4-mini-instruct (3.8B) fine-tuned with QLoRA on the QuerySmith-Spider-Bird dataset (derived from Spider and BIRD), then compressed with a partial-residual quantization scheme and served with llama.cpp. The final model reaches 0.743 exact-match accuracy at 22.3 ms/token and 3.5 GB, against an FP32 baseline of 0.713 at 148 ms/token and 14.9 GB (6.6x faster, 4.2x smaller).
LLM & RAG · 2026
RAG research assistant: users upload PDFs, embeddings are stored and indexed, relevant chunks are retrieved by semantic search, and the LLM answers with citations. Ships with REST APIs and Swagger docs.
LLM & RAG · 2026
Production-ready PDF question answering: ingests documents, stores vector embeddings, retrieves relevant chunks and generates document-grounded answers.
LLM & RAG · 2026
Grounded Q&A over study material using FastAPI, LangChain, FAISS, Hugging Face embeddings and Mistral AI.
LLM & RAG · 2026
Semantic resume search and candidate ranking against job descriptions using Mistral embeddings and FAISS, going beyond keyword matching.
LLM & RAG · 2026
LLM chatbot (Mistral-7B-Instruct via Hugging Face Inference API) with prompt engineering and safety filters that give non-diagnostic health guidance and block prescription or emergency queries. Real-time token streaming and session chat history in Streamlit.
Full-Stack · 2026
AI-powered system built around Laya, a structured decision model for fast probabilistic decisions. Instead of calling a generative LLM for every task, Laya turns an unstructured request into structured decisions (category, priority, risk, escalation probability). A deterministic policy engine interprets them and a workflow engine acts (assign, escalate or queue). Example: "Our production database is unavailable and all API requests are failing" becomes Category Technical, Priority Critical, Escalation 0.94, routed to the Engineering team and escalated. Components: Laya decision layer, decision normalizer, policy engine, workflow engine, FastAPI backend, persistence layer, testing layer. Goal: a small specialized model for structured decisions, deterministic software for business rules, and large generative models only when language generation is really needed.
Full-Stack · 2026
Full-stack platform doing syntax validation, bug detection, security scanning, complexity analysis, optimization suggestions and automatic documentation, through a six-stage sequential analysis pipeline with real-time NDJSON streaming.
Full-Stack · 2026
End-to-end employee burnout prediction with a stacking ensemble (Random Forest, Gradient Boosting, Decision Tree), a FastAPI inference API and a React + TypeScript frontend; models serialized with Joblib.
Full-Stack · 2026
React + Vite storefront with a Django REST backend: animated hero, parallax scroll, product grid, slide-in cart drawer, quick view, multi-step checkout, GSAP ScrollTrigger and Framer Motion.
Full-Stack · 2026
FastAPI + Streamlit task system with CRUD, search, priority and completion filters, SQLAlchemy ORM with SQLite, Pydantic v2 schemas.
Full-Stack · 2026
FastAPI service that retrieves historical and current stock data and returns analysis.
Full-Stack · 2026
Logistic Regression model chosen with GridSearchCV (Accuracy, F1, AUC) and deployed as a FastAPI service.
Deep Learning · 2026
Interactive Python app that converts mathematical equations such as sin(x), x² or sin(x) + cos(2x) into sound: the user picks a math-to-audio mapping and generates the waveform. Covers sampling, frequency, amplitude, phase, PCM, WAV generation, FFT and spectrograms. Built as a learning project with a future path toward audio ML and mathematical function recognition.
Deep Learning · 2026
Learns dense embeddings from movie metadata with a Siamese Transformer and contrastive loss (weak-supervision pairs), indexed in FAISS for millisecond similarity search, served with FastAPI.
Deep Learning · 2026
Neural matrix factorization combining collaborative and content signals, FastAPI + PostgreSQL.
Deep Learning · 2026
Classifies 50 sound categories (ESC-50) by converting waveforms to log-Mel spectrograms and using a 2D CNN in PyTorch.
Deep Learning · 2026
End-to-end NLP pipeline in PyTorch classifying news into World, Sports, Business and Sci/Tech: preprocessing, vocabulary, training, evaluation, inference and a REST API.
Deep Learning · 2026
Bidirectional LSTM in PyTorch classifying text into six emotions.
Deep Learning · 2026
Deep learning pneumonia detection from chest X-rays.
Deep Learning · 2026
Real-time mask detection with PyTorch, MobileNetV2 transfer learning and OpenCV, webcam inference and a modular pipeline.
Deep Learning · 2026
Hybrid CNN-LSTM on raw smartphone accelerometer and gyroscope signals with subject-wise validation and MLflow tracking.
Deep Learning · 2026
CNN training, evaluation, REST inference, structured logging and automated tests.
Deep Learning · 2026
OpenCV Haar Cascade face detection from images, video and webcam through FastAPI.
Deep Learning · 2025
Draw a digit and get real-time CNN predictions.
Deep Learning · 2026
Fine-tuned BERT for multi-class news classification.
Deep Learning · 2026
Forward/backpropagation and gradient descent in pure NumPy.
Deep Learning · 2026
LSTM text generator trained on famous quotes.
ML & Data · 2026
Predicts which customers are likely to churn, served as an interactive Streamlit app.
ML & Data · 2026
Regression model predicting student exam scores, with a Streamlit interface.
ML & Data · 2026
Compares eight regression models (R² and error) and serves the best in a dashboard.
ML & Data · 2026
Mall Customers dataset, income vs spending score.
ML & Data · 2026
Hugging Face Transformer sentiment analysis with real-time, CSV batch and demo modes.
ML & Data · 2025
Lightweight sentiment analysis app for quick text checks.
ML & Data · 2025
ARIMA and Prophet with Plotly and CSV export.
ML & Data · 2025
Automated EDA, cleaning and visualization.
ML & Data · 2025
SVM, Decision Tree and KNN comparison.
ML & Data · 2025
OpenCV filters (Gray, Canny, Blur, Sketch) in Streamlit.
IoT · 2026
ESP32, HC-SR04 ultrasonic sensor and servo that sweeps 180 degrees; an LED triggers when an object is within range; distance and angle printed to the serial monitor.