> ls ~/projects

Selected Work

Seven projects across machine learning, computer vision, data engineering, agentic AI, quantitative finance and NLP. Each one is a thing I built and broke and learned from — open any card for the full case study.

emotionflow.md
JOY · 0.82WEBCAMAUDIOnow matchingspotify
01

EmotionFlow

ACTIVE

ML · Real-time Vision · Affect Systems · 2026

An affect-aware music system built around a four-pipeline architecture — emotion detection, state control, queue management, and Spotify integration — engineered to stay stable under the noise of real-world webcam input.

  • Python
  • PyTorch
  • OpenCV
  • PyQt6
  • Spotipy
read the case study
risk-surrogates.md
10³10⁴10⁵10⁶10⁷SAMPLES →VaR / CVaRMonte Carlo (100M)Surrogate~1% error · 1000x faster
02

Monte-Carlo Risk Surrogates

SHIPPED

Deep Learning · Quant Finance · 2025

Gaussian-process and neural-network surrogates that estimate portfolio VaR / CVaR in milliseconds — within 1% of a 100M-sample Monte Carlo baseline.

  • Python
  • PyTorch
  • Gaussian Processes
  • NumPy
read the case study
oceaniq.md
CMLREraw · csvAngria Bankraw · csvMarine Surveysraw · csvETLclean · map · mergeunified.schema
03

OceanIQ

MVP

AI Data Platform · Pipeline · 2025

An AI-driven data pipeline that unifies fragmented Indian oceanographic datasets — CMLRE, Angria Bank, marine-mammal surveys — into one clean, analysis-ready schema.

  • Python
  • Jupyter
  • Pandas
  • SQL
read the case study
explainmyrepo.md
github.com/blazingshadows/repo▸ explain▸ src/main.pymodel.pyutils.py▸ tests/README.mdLLMWalkthrough
04

ExplainMyRepo

ACTIVE

LLM Tooling · Developer UX · 2025

Point it at any GitHub repository and it generates a human-readable walkthrough — file structure, key components, and overall workflow — using AI.

  • Python
  • LLMs
  • GitHub API
read the case study
sentinel-capital.md
9 agents · NSE/BSE · paper tradingmarket dataBreeze APIMoneyControl RSS229 NSE symbolsTechnicalpure mathNews & Sent.GeminiMacroOpenAIForecastingLightGBMContrarianAnthropic · debatedamps contradicted agentsDCS consensusinfluence × trust × contextDCS threshold: 0.15GARCH vol guardExecutiondelta-sizedSQLite auditFULL AUDIT TRAIL · EVERY CYCLE · EVERY NO-TRADE
05

Sentinel Capital

ACTIVE

Multi-Agent System · Quant Finance · Agentic AI · 2026

A 9-agent autonomous investment committee for intraday NSE/BSE paper trading. Five specialist AI agents debate each signal through a Dynamic Trust Framework — GARCH(1,1) volatility guards, Ledoit-Wolf shrinkage diversification, full trade explainability enforced.

  • Python
  • FastAPI
  • LightGBM
  • LangGraph
  • SQLAlchemy
  • TypeScript
  • Gemini
  • OpenAI
  • Anthropic
read the case study
skeptitix.md
PDFSection Parserfont-size awareLangGraphClaim AgentThreadPoolExecutorEvidence Agentvector searchMemory AgentChromaDB readSkeptic AgentLimitations + Questionsno verdictsReport AgentChromaDB · 3 collections
06

SkepTiX

SHIPPED

Multi-Agent NLP · Research Analysis · Agentic AI · 2026

A five-agent LangGraph system that answers "why should I believe this paper?" — extracting claims, mapping evidence, surfacing limitations, and generating a structured skepticism report from any academic PDF.

  • Python
  • LangGraph
  • Gemini 2.5 Flash
  • ChromaDB
  • FastAPI
  • Next.js
  • sentence-transformers
read the case study
network-congestion-prediction.md
12345678Abilene 12-node topologyGATConv ×2spatial encodingLSTMT=20 sequenceMulti-Head Attnresidual fusion3-class outputNormal75%Warning62%Congested38%77.13% acc · 0.8973 AUC/ROC
07

Network Congestion Prediction

SHIPPED

Graph Neural Networks · Spatiotemporal ML · 2026

Three progressively deeper GNN architectures — GAT, GAT+LSTM, GAT+LSTM+Attention — for predicting congestion classes on the 12-node Abilene backbone, with a focus on diagnosing and fixing training bottlenecks.

  • Python
  • PyTorch
  • PyTorch Geometric
  • GATConv
  • LSTM
  • scikit-learn
  • NumPy
read the case study
Aditya Dixit · Jaipur, India7 of 7 shown© 2026