Projects

2026

Feb. 2026 – Present

Engineered agentic inference pipeline on Modal H100s for real-time 3D spatial reasoning without LiDAR
Optimized object re-detection using SAM3 and CLIP for real-time performance on cloud infrastructure
Designed retroactive re-detection enabling dynamic scene updates and 3D reasoning over mapped environments

Python Modal H100s VGGT

GitHub

Jan. 2026 – Mar. 2026

Built interactive SaaS analytics dashboard analyzing ARR waterfalls and retention risk metrics
Integrated multiple LLM providers for automated churn detection and revenue impact scoring
Deployed containerized BI solution with Docker enabling real-time customer success insights

Python Streamlit Plotly Docker

GitHub

May 2026 – Present

Engineered localized SQLite indexing pipelines to parse and query historical university data via relational schemas
Optimized backend data aggregation to resolve cross-listed rubrics and filter post-2021 grade spikes globally
Designed interactive asynchronous UI to dynamically map multi-category Gen-Ed performance statistics

Python Discord.py pandas Matplotlib

GitHub

2025

Dec. 2025 – Present

Generated synthesizable microbial genomes from natural language via LoRA fine-tuned Evo 1/2 foundation model
Built 4-node LangGraph agentic pipeline for constraint-aware strain planning over KEGG/MGnify genomic data
Trained 4-layer GAT on 20K+ microbial interaction graphs for 90-day community stability prediction

Evo 2 FastAPI LangGraph PyTorch Geometric Next.js

GitHub

Oct. 2025 – Nov. 2025

Designed an AI-powered chatbot for predicting CRISPR screen results using Ollama 3.1 finetuned RAG
Integrated semantic search (all-MiniLM-L6-v2 embeddings) and top-document retrieval for fast response times
Utilized geospatial, historical, and literature-retrieved health data to enhance pathway analysis for CRISPR

Python NumPy pandas PyTorch

GitHub

Aug. 2025 – Present

Created JIT-compiled library generating 100K+ Monte Carlo paths in approx. 2s on consumer CPU
Implemented Black-Scholes, Heston stochastic volatility, and GARCH models with 90% test coverage
Supports Iron Condor/Butterfly Spread strategies optimized via Numba for 100K+ paths

Python NumPy pandas ARCH Numba

GitHub