Archive
Selected systems, models and agents — each entry documents what the thing does, how it was built, and what it turned out to be good for.
- ENTRIES
- 009
- RANGE
- 2023–2025
- DOMAINS
- 008
- STACK
- 030
Sentic Quant
Quantitative Finance
A multi-modal signal engine fusing FinBERT sentiment with a Time-Series Transformer architecture. It decodes non-linear market regimes by applying self-attention to unified qualitative and quantitative data. Features a robust walk-forward backtesting suite.
VectorLoom
NLP / Generative AI
A decoupled FastAPI/Streamlit architecture utilizing local embedding models and language generation to weave retrieved document chunks into highly accurate, domain-adaptable answers.
Verity OS
Agentic Systems
A multi-agent research engine that fact-checks claims and curates niche expert resources. Agents divide retrieval, verification, and synthesis so every conclusion traces back to a source rather than a single model's assertion.
Multimodal Instructional Assistant
Multimodal AI
A multimodal AI pipeline for educational content understanding and contextual question generation. It reads lecture video, slides, and transcript together, then generates questions grounded in what was actually taught.
VoiceForge
Voice AI Agent
A voice-controlled local AI agent that transcribes audio, classifies intent, and executes actions on your machine — powered by Whisper, Ollama, and Streamlit. Everything runs on-device, so nothing leaves the machine.
Social to Lead
Agentic Workflow
A social-to-lead agentic workflow built with LangGraph and Groq (Llama 3.3 70B). The agent classifies intent, answers product and pricing questions from a RAG knowledge base, and collects lead information through natural conversation before writing qualified leads to a CRM.
Multi-Asset Correlation Heatmap
Quantitative Finance
A portfolio analysis tool that detects shifting market regimes and maps how asset correlations break down under stress — the moments when diversification quietly stops working.
Brain Tumor Detection
Computer Vision
A compact, reproducible pipeline for detecting and classifying brain tumors from MRI scans using transfer learning on MobileNetV2. Covers data organization, augmentation, training, checkpointing, evaluation, and a minimal Streamlit dashboard.
Energy Demand Forecasting
Predictive Modeling
Time-series forecasting of hourly energy consumption using an LSTM network, trained on the AEP_MW load dataset to predict demand curves ahead of the grid.
