ML Engineer building production systems at the intersection of machine learning, quantitative finance, and AI governance. I contribute to open-source projects at Google, Microsoft, NVIDIA, and MONAI, and build tools that make ML in regulated industries less painful.
class AtharvaJoshi:
def __init__(self):
self.location = "New York, USA"
self.current_focus = ["AI Governance", "Quantitative Finance", "Agent Evaluation"]
self.interests = ["Regulated ML", "Market Microstructure", "LLM Safety"]
self.languages = ["Python", "C++", "SQL", "TypeScript"]
def say_hi(self):
print("Let's build something amazing together!")
me = AtharvaJoshi()
me.say_hi()| Organization | Contribution |
|---|---|
| Microsoft | EU AI Act risk classifier for agent-governance-toolkit (merged) |
| Security fix in tf-quant-finance (PR open) | |
| MONAI (NVIDIA) | Gaussian kernel fix in medical imaging loss functions (PR open) |
| FinRL | Threading bug fix in paper trading engine (merged, 14.6k stars) |
| Goldman Sachs | Pandas 2.x compatibility for gs-quant (PR open) |
| sktime | NaiveForecaster bug fix (PR open, 9.7k stars) |
Train a model, get a compliance report. Built for teams shipping ML in regulated industries.
- Wraps any scikit-learn estimator with automatic governance
- SHAP explanations, fairness audits (demographic parity, disparate impact)
- EU AI Act compliance checks (Articles 5, 10, 12, 13, 14, 15)
- Data drift detection (Kolmogorov-Smirnov, PSI)
- Auto-generated model cards and tamper-proof audit logs
Tech: Python · scikit-learn · SHAP · Pydantic
Measure accuracy, cost, latency, and safety across any AI agent architecture.
- Works with any agent: LangChain, CrewAI, AutoGen, or plain functions
- Safety checks: PII leak detection, prompt injection, custom patterns
- Async runner with configurable concurrency
- Side-by-side agent comparison with winner selection
Tech: Python · Pydantic · tiktoken · asyncio
Sub-microsecond order book operations with ML-powered trade flow prediction.
Tech: C++20 · Python · XGBoost · Lock-free Programming
Deep Galerkin Method for solving Black-Scholes and exotic option PDEs.
Tech: PyTorch · Neural Networks · Quantitative Finance
Predicts cash crises 90 days before they hit. Full-stack with CI/CD.
Tech: Python · FastAPI · React · Docker · GitHub Actions
Multi-agent simulation for predicting outcomes through collective behavior.
Tech: Python · Multi-Agent Systems · Simulation
Languages
ML & Data Science
Infrastructure & Tools
