Hello, I'm
Purna Chander Konda
I architect production-ready AI systems that transform complex data into measurable business outcomes — driving innovation across finance, healthcare, and retail.
About Me
My professional background and qualifications
Results-driven AI/ML Engineer with 4+ years of experience designing and deploying production-ready ML models and scalable AI systems across finance, retail, and healthcare.
Expert in Generative AI, RAG, and MLOps using Python, PyTorch, TensorFlow, Hugging Face, LangChain, and AWS. Skilled in NLP, Computer Vision, and Predictive Analytics with a proven track record of automating workflows, improving efficiency, and delivering secure, high-impact business outcomes.
Committed to Responsible AI and regulatory compliance — building AI systems that are transparent, fair, and auditable.
Master's in IT Management
Webster University, MO, USA
Bachelor's in CS & IT
CVR College of Engineering, India
AWS Certified Data Engineer
Associate
Azure AI Engineer Associate
Microsoft Certified
My Skills
Comprehensive expertise across the AI/ML technology stack
Machine Learning & NLP
MLOps & Deployment
Responsible AI
Programming & Tools
Data Analytics
My Projects
Showcasing impactful AI/ML solutions I've built
Deepfake Detection Using SWIN Transformer
Fine-tuned SWIN-Tiny (microsoft/swin-tiny-patch4-window7-224) on 190,000+ real and fake facial images to classify deepfakes with 98.81% accuracy. Built a hierarchical vision transformer pipeline with shifted-window multi-head self-attention for local and global feature extraction. Deployed as a live Gradio web app on Hugging Face Spaces with real-time confidence scores — no sign-up required.
Music Genre Classification — CNN & Random Forest
End-to-end audio classification pipeline categorising music tracks into 10 genres (Blues, Classical, Country, Hip-Hop, Metal, and more) across three model architectures: CNN on raw mel-spectrograms, CNN on feature-engineered spectrograms, and a Random Forest baseline. Extracted MFCCs, spectral centroids, and chroma features using Librosa; deployed interactively via a Gradio web interface with full Google Colab reproducibility notebooks.
RAG-Based PDF Research Assistant
Production-grade Retrieval-Augmented Generation system that lets you upload any PDF and chat with it in natural language. Chunks documents via RecursiveCharacterTextSplitter, embeds with BAAI/bge-small-en-v1.5 SentenceTransformers, and stores vectors in a FAISS index for sub-second semantic retrieval. Supports fully offline inference via Ollama (Mistral 7B / LLaMA / Phi) or cloud via Gemini. Ships with a benchmarked evaluation pipeline — achieving 35% higher recall over keyword search at ~1.2s average query latency. Docker-ready and deployed on Streamlit Cloud.
NexaChat AI — Conversational Assistant with 3D Avatar
Full-stack conversational AI web app featuring a 3D interactive robot avatar rendered with Babylon.js, real-time AI chat responses, and a sleek dark UI with smooth animations. Engineered a multi-model fallback chain cycling through Mistral Small 3.1 24B → Qwen3 4B → Gemma 3 4B via OpenRouter API for maximum uptime — gracefully handling provider outages without user disruption. Backend built with Flask + Gunicorn, deployed and production-hardened on Render with environment-aware port binding.
My Experience
Professional journey across top organizations
AI/ML Engineer
Broadridge
- Fine-tuned foundation models (LLaMA, GPT) using PEFT, LoRA, and RLHF; implemented Chain-of-Thought reasoning to boost compliance metrics by 30%.
- Architected scalable RAG-based AI pipelines using LangChain and Pinecone, enhancing document processing efficiency by 40% across large-scale vector stores.
- Deployed high-performance AI microservices via AWS Bedrock and gRPC, supporting real-time financial workflows for 15,000+ enterprise users.
- Streamlined production MLOps using Kubeflow and SageMaker Pipelines, achieving 99.9% system reliability and accelerating deployment cycles by 40%.
- Optimized model inference performance with TensorRT and GGUF quantization, slashing latency by 35% for mission-critical financial applications.
- Engineered high-concurrency fraud detection pipelines with Spark and Vertex AI, increasing anomaly detection rates by 25% for financial transactions.
- Built Explainable AI (XAI) frameworks using SHAP and LIME, ensuring model transparency and strict adherence to GDPR, FINRA, and FCRA regulations.
- Developed sophisticated financial Knowledge Graphs in Neo4j, enabling analysts to extract risk-related insights 50% faster than traditional methods.
Machine Learning Engineer
O'Reilly Auto Parts
- Architected robust ETL and ML pipelines utilizing Spark, Airflow, and Snowflake to orchestrate 100TB+ of transactional data for enterprise-scale analytics.
- Developed high-precision demand forecasting models using XGBoost and Prophet, cutting stockout occurrences by 18% across 1,200 distribution hubs.
- Deployed containerized inference services via Docker and Kubernetes, achieving sub-200ms latency for real-time pricing optimization models.
- Leveraged BERT-based NLP architectures to classify 500,000+ unstructured parts, contributing to a 10% decrease in overstock.
- Optimized MLOps workflows using AWS SageMaker and MLflow, resulting in a 35% reduction in model retraining cycles.
- Built real-time streaming pipelines with Kafka and Flink to enable proactive anomaly detection across 28 logistics centers.
- Established automated CI/CD pipelines via GitLab and GitHub Actions, ensuring 99.9% system availability for mission-critical workloads.
- Engineered operational dashboards using Grafana and Tableau, accelerating executive decision-making processes by 30%.
Data Scientist
Sahrudaya Healthcare
- Fine-tuned BERT and RoBERTa architectures via Hugging Face to classify 1M+ medical records, achieving a 28% improvement in diagnostic classification accuracy.
- Developed predictive risk models using XGBoost and LightGBM, increasing patient engagement by 22% and reducing hospital readmissions by 15%.
- Architected HIPAA-compliant ETL workflows using Spark and Hive on Azure/GCP, ensuring 99.9% data reliability for sensitive medical datasets.
- Engineered Computer Vision pipelines with CNNs and OpenCV for medical imaging analysis, accelerating automated screening speeds by 30%.
- Constructed time-series forecasting models using LSTM and Keras to predict ER capacity, improving patient throughput by 20%.
- Developed model explainability (XAI) frameworks using SHAP and MLflow, integrating automated drift detection and A/B testing protocols.
Data Scientist
Citco
- Architected GNN-based fraud detection models and link analysis pipelines using Neo4j, reducing false-positive rates by 27% through advanced relationship mapping.
- Engineered scalable enterprise pipelines via Databricks and Spark MLlib, cutting operational reporting turnaround by 40% for financial stakeholders.
- Developed high-performance time-series ensembles using CatBoost and LSTM, significantly increasing underwriting accuracy and risk assessment.
- Optimized distributed query performance in Cassandra and MongoDB, reducing data retrieval latency by 40% for high-frequency access patterns.
- Deployed production ML microservices using Docker and Kubernetes, seamlessly integrating Tableau and Power BI for real-time visual analytics.
Get In Touch
Let's collaborate and build something impactful
I'm always open to discussing AI/ML projects, consulting opportunities, or full-time roles. Feel free to reach out!
Chat on WhatsApp