Applied AI Engineer. I build and ship LLM systems end to end: RAG pipelines, agent orchestration (MCP), LLM evaluation frameworks, and ML models with explainability. MS in Information Systems, Northeastern (2026). STEM OPT, no sponsorship required.
I care about the unglamorous parts that make AI actually work in production: evaluation harnesses, deployment gating, latency budgets, and catching the data-leakage bug before it ships an AUC of 1.0.
- LLM evaluation: regression testing against golden examples, structural constraint validation, retrieval-confidence monitoring, automated deploy gates
- RAG: Hugging Face embeddings, FAISS vector search, cosine-similarity thresholding
- Agents: tool-calling agents and orchestration via the Model Context Protocol (MCP)
- ML with explainability: XGBoost / LightGBM, SHAP attribution, bias auditing, human-in-the-loop workflows
PredictiveCare: HIPAA-compliant clinical risk prediction
XGBoost + LightGBM on 215M+ clinical events across 200K+ patients. Temporal train/test splits after catching a data-leakage bug. Per-patient SHAP attribution, disparate-impact bias auditing, and human-in-the-loop clinical workflows.
Python FastAPI PostgreSQL XGBoost SHAP Next.js
AI Playlist Mixer: real-time multi-device music coordination
Unifies Spotify and YouTube into a single queue synced across up to 7 devices, with fairness-aware reranking and AI-generated session summaries. Live demo
FastAPI Next.js Spotify API YouTube API Railway Vercel
Languages: Python · Java · C++ · TypeScript · SQL ML/AI: PyTorch · TensorFlow · scikit-learn · XGBoost · LightGBM · SHAP · Hugging Face · FAISS GenAI: RAG · MCP · Claude API · OpenAI API · AWS Bedrock · agent orchestration Backend/Cloud: FastAPI · Flask · Docker · CI/CD · AWS (Bedrock, S3, Lambda) · PostgreSQL · MongoDB
- Email: joshi.sus@northeastern.edu
- LinkedIn: https://www.linkedin.com/in/sushant-joshi44/
- Portfolio: https://sushant-joshi-mu.vercel.app/