🎓 CS & Applied Math @ NYU • Data Science · AI/ML

B2B AI platform that delivers deal analysis and consulting insights to SMBs.

• Built multi-agent orchestration system with RAG pipelines for document analysis
• Designed vector search layer (pgvector + Supabase) handling structured and unstructured data
• Optimized inference costs 40% via model routing (GPT-4o for complex queries, DeepSeek for routine tasks)

Tech: FastAPI, PostgreSQL, Supabase, pgvector, LangChain

CSV to tested metrics + dashboards + plain-English insights.

Repo · Demo

• Cleans common data issues (missingness, duplicates, schema drift)
• Computes metric layer: cohorts, trends, segments
• Routes questions via SetFit classifier to appropriate analysis pipeline

Tech: FastAPI, Postgres, Redis, Streamlit, dbt, Prefect

Business question → structured report (JSON + deck + spreadsheet).

Repo

• Converts vague prompts into structured analysis steps
• Retrieval-grounded generation (outputs tied to source data)
• Benchmarked cost/latency tradeoffs (results in /benchmarks)

Tech: LangGraph, FastAPI, Redis, Docker

Framework to distinguish real trading signal from noise.

Repo · Demo

• Deterministic backtests with transaction cost modeling
• Statistical validation: bootstrap CI, permutation tests, Monte Carlo simulation
• Clear decision rules: edge vs inconclusive vs noise

Tech: Python, pandas, NumPy, SciPy, Streamlit, pytest

NLP + hypothesis testing on lyrical style evolution.

Repo · Demo

• Sentiment analysis, lexical diversity metrics, topic modeling
• Statistical tests across albums (not just visualization)

Tech: Python, scikit-learn, React, Vite

Languages: Python, SQL
Data: pandas, NumPy, SciPy, scikit-learn, dbt
Infrastructure: FastAPI, PostgreSQL, Redis, Docker, Prefect
ML/AI: LangChain, LangGraph, RAG pipelines, vector databases

💡 I like building things that actually work, then figuring out why they work.