👩‍💻  Palla Chetana Reddy     AI Engineer · ML Engineer · Data Scientist · Founder

📍  Hyderabad, India 🇮🇳

🎓  B.Tech CSE — Data Science  2023 – 2027     Woxsen University  ·  🥇 Gold Medalist

🔨  Currently Building     🛡️ Shadow AI — Enterprise LLM Security Gateway     🏢 Oronzo LLP — Multi-vertical AI tech group

🌐  Domains     Healthcare AI  ·  LLM Security  ·  Agentic AI     Causal ML  ·  Computer Vision  ·  Fintech     Cybersecurity  ·  NLP  ·  Supply Chain

💭  Beliefs     › Real impact > benchmark chasing     › Security + AI = non-negotiable     › Build it. Ship it. Measure it.

⭐  What Drives Me     › Digging up problems no one's solved yet     › Curiosity to search · learn · fail · progress     › AI/ML · Data Science · Explorer at heart

Enterprise AI Security Gateway — Currently Building

Production-grade gateway between users and any LLM. Intercepts every prompt, scans for threats, enforces per-org policies — all in real time.

• 🔍 15 security layers: Auth (JWT+SHA-256) → Rate Limiting → 20+ Regex patterns → Semantic Injection (TF-IDF + LogReg) → PII Scanner → Topic Classification → Policy Engine → Response Scanner
• 🧠 IsolationForest per-user behavioural anomaly detection
• 🗃️ PostgreSQL audit log — every interaction persisted
• 🔔 Slack + SMTP alerts on critical risk
• 📊 6-tab Streamlit SOC dashboard — threats, topics, user behaviour, time-series
• 📈 Prometheus /metrics endpoint + Admin API

FastAPI IsolationForest TF-IDF PostgreSQL Docker Prometheus Streamlit

AI Agent Incident Response Benchmark

OpenEnv benchmark for AI agents diagnosing & resolving fintech production outages.

• 🎯 First agent run scored 0.93 / 1.0
• 🔁 5 difficulty-graded scenarios, 13 typed actions
• 🧪 90 tests, all passing — Dockerized on HF Spaces
• 🔌 Any LLM or RL agent pluggable via REST

OpenEnv FastAPI Docker Pydantic pytest

AI-Powered Cloud Honeypot & SOC

• 🐝 Adaptive SSH + Web honeypots evolving on attacker behavior
• 🌲 Random Forest + Isolation Forest on CIC-Honeynet
• 📊 Live React SOC dashboard: attack maps, heatmaps
• ⚡ Full incident lifecycle management

Python FastAPI React Random Forest Docker

ICU Causal ML & Counterfactual Dashboard

Per-patient treatment effect estimation on 500 real ICU patients × 383K vitals.

• 🎯 X-Learner: ATE bias 0.02 days, CATE r = 0.77
• 🛡️ E-value = 3.54 — robust to hidden confounders
• 📊 6-tab what-if simulator: slide age/SpO₂ → live CATE
• 🏥 Built during Apollo Hospitals internship

CausalML XGBoost SHAP Meta-Learners Streamlit

Real-Time Air Writing Recognition

Write characters in the air with your finger — recognised on screen in real time.

• 🖐️ MediaPipe Hands — 21 landmarks at 30+ FPS
• 🧠 Custom CNN on EMNIST — ~88% accuracy, 47 classes
• ✋ 7 intuitive gestures with EMA stroke smoothing
• 🌐 Streamlit + WebRTC browser version included

MediaPipe OpenCV CNN EMNIST PyTorch Streamlit

Explainable AI Medical Imaging

• 🧠 CNNs + Swin Transformer on 5,000+ medical images
• 🗺️ Grad-CAM explainability for clinical interpretability
• 👨‍⚕️ Human-in-the-loop validation reducing false positives

PyTorch Swin Transformer Grad-CAM Computer Vision

Multimodal Soft Skill Detection

Audio + text fusion model for real-time communication skill analysis.

• 🎤 Audio features via librosa + text NLP fusion
• 🧠 TensorFlow/Keras multimodal classification
• 🌐 Flask web app deployment

TensorFlow Keras librosa OpenCV Flask

Self-Healing Drone Swarm Network

Multi-agent RL for adaptive drone communication networks with Unity 3D visualisation.

• 🌐 96.8% network stability, avg recovery: 4 steps
• 🧠 GAE + VGAE + Graph GAN + Graph Diffusion
• 🕹️ Unity 3D swarm formation demo
• 🤖 Decentralized MARL control

PyTorch GNN GAE VGAE MARL Unity 3D

Regime-Aware NIFTY 50 Forecasting

1-day, 7-day & 30-day horizons with dynamic ensemble switching.

• 🧩 LSTM + TFT + Prophet — optimal per market regime
• 📊 Monte Carlo Dropout + Quantile Regression bands
• ✅ Sub-1% MAPE on global ETF cross-validation
• 🔁 Live backtester: any date 2015–2024, no look-ahead

LSTM TFT Prophet Regime Detection VaR CVaR

Distributed Fraud Detection

• 🔥 PySpark pipeline — 6M+ transactions, AUC-ROC 0.9995
• 🌊 Kafka + Spark Structured Streaming — 50 events/sec
• 📊 Low/Medium/High risk scoring (0–100)

PySpark Kafka GBT Docker Streamlit

Clinical Analytics Pipeline

• ⏱️ 61,930 records ingested into InfluxDB
• 🚨 Automated anomaly + deterioration detection
• 📦 Published 49.7 MB dataset on Kaggle

InfluxDB SQL Python Time-Series

Predictive Risk Management

• 📈 28% disruption detection improvement
• 🗄️ Star Schema — 18+ datasets integrated
• 🗺️ 1,000+ route graph simulations (NetworkX)

XGBoost LightGBM GNN Power BI

Offline Agentic AI Chatbot

• 🎯 95% response accuracy, 10+ intent types
• 💰 100% cost reduction — fully offline Ollama LLM
• 🔗 LangGraph multi-turn memory management

LangGraph LangChain Ollama (Phi-3) NLP

🎓 Academic Excellence

🏆 Competitions & Innovation

🌏 International & Recognition

🎯 Olympiads & Early Excellence