CS Student · ML · iOS · Backend

Aditya
Bhatt.

B.Tech CSE (Data Science) · PSIT Kanpur · Expected June 2027.
3rd year CS student who learns by building. These are the projects I've shipped so far.

GitHub LinkedIn Email

// Projects

ProteinLens On-Device ML · iOS

Real-time food classification running fully on-device via CoreML. Point your camera at food — instant nutrition estimates, zero network calls, fully private.

  • MobileNetV2 fine-tuned on 20-class India-optimised Food-101 — 86% val accuracy
  • CoreML `.mlpackage` (5.9 MB) via TensorFlow → SavedModel → coremltools pipeline
  • AVFoundation + Vision + CoreML live inference — < 50ms on Neural Engine
  • Serving size slider (50g–500g), daily scan history, freeze-frame tap-to-lock
SwiftCoreMLVision AVFoundationTensorFlowMobileNetV2coremltools
GitHub →
FitMacroAI Full-Stack iOS · Live

AI-powered iOS fitness app. Type natural language food input in Hindi or English — get instant, accurate macro breakdowns backed by a live AI backend.

  • Type "rajma chawal 1 katori" → 420 cal, 19g protein, 60g carbs, 14g fat
  • LLaMA 3.3 70B (Groq) for food parsing; FastAPI backend live on Railway
  • SwiftUI — onboarding flow, daily auto-reset, UserDefaults persistence
SwiftUIFastAPILLaMA 3.3 70B Groq SDKRailwayPython
iOS App → Backend → Live API →
DocuMind RAG · AI · Live

AI document Q&A with strict no-hallucination guarantees. Upload any PDF, ask in plain English, get answers with exact page citations.

  • Full RAG pipeline: HuggingFace embeddings → ChromaDB → LLaMA3-70b (Groq)
  • Confidence indicator (green/red), auto-summary on upload, document-only prompting
  • FastAPI backend + Streamlit frontend — live at docu-mind03.streamlit.app
LangChainChromaDBFastAPI StreamlitHuggingFaceGroq
GitHub → Live Demo →
Credit Risk Analyzer ML · Financial AI

Alternative credit scoring system trained on 307,511 real loan applications. FICO-style score output, FastAPI-ready.

  • Logistic Regression AUC 0.6063 → XGBoost AUC 0.6805 (+12.3% lift)
  • Feature engineering across 300k+ records from raw financial indicators
  • Score output: 300–900 (FICO-style)
XGBoostScikit-learnPandas NumPyPython
GitHub →

// Tech Stack

ML / Data

Python · TensorFlow · CoreML
coremltools · XGBoost
Scikit-learn · Pandas · NumPy

iOS

Swift · SwiftUI · AVFoundation
Vision · CoreML
iOS 17+

Backend

FastAPI · REST APIs
Railway · LangChain
ChromaDB · Streamlit

AI / LLM

Groq SDK · LLaMA 3.3 70B
HuggingFace · Prompt Eng.
RAG pipelines