Praveen Kumar

BharatOne, Pune

• Built and shipped a production chat application (Firebase Firestore/Auth/Cloud Functions) with end-to-end encrypted messaging and a caching layer, serving real users with low-latency delivery and native social-media sharing via a JWT-secured backend.
• Designed and integrated an AI chatbot with custom guardrails to sanitize inputs/outputs and block prompt-injection and abuse attempts, hardening the assistant against adversarial attacks in a live production environment.

Aigetai, Bengaluru

• Architected and deployed production-grade Retrieval-Augmented Generation (RAG) pipelines, integrating advanced AI gateways and custom routing logic for optimized LLM inference.
• Engineered robust guardrails and comprehensive evaluation (evals) frameworks to ensure output safety, reliability, and alignment with enterprise quality standards.

Outlier, United States

• Improved LLM output quality across STEM domains through advanced prompt engineering, chain-of-thought reasoning, and RLHF evaluation workflows, ensuring technical accuracy and ethical alignment.
• Executed Reinforcement Learning from Human Feedback (RLHF) workflows to evaluate and refine model outputs across coding and scientific reasoning tasks.

TrendalyTix

• Architected predictive models using Scikit-learn and Pandas to analyze real-world datasets, improving data-driven decision accuracy by 25%.
• Optimized automated data cleaning pipelines for production-grade analytics workflows, reducing preprocessing time by 40%.
• Synthesized complex technical findings into actionable business insights, presenting stakeholders with interactive visualizations using Matplotlib and Seaborn.

• Built a Continuous Temporal Consistency & Faithfulness Evaluator for Vision-Language Models in real-time video streams, ensuring output reliability at scale.
• Designed guardrail pipelines that detect hallucinations and inconsistencies across multimodal VLM outputs, reducing unsafe model responses in production environments.
• Implemented real-time evaluation framework processing video streams with sub-second latency for enterprise-grade AI safety compliance.

• Developed an autonomous Cognitive Agent using LangGraph that plans analysis steps, executes Python code, and autonomously resolves runtime errors via recursive reflection loops (Self-Healing).
• Deployed a secure, containerized application using Docker on Render with a Streamlit interface, implementing sandbox guardrails for safe code execution.

• Engineered a Multi-Agent GraphRAG system that extracts biomedical entities from unstructured clinical PDFs and constructs a Neo4j knowledge graph for clinical trial matching.
• Implemented agentic multi-hop reasoning to match patients with eligible trials using cited evidence, achieving accurate entity extraction across Drug-Protein-Disease relationships.
• Built an end-to-end pipeline combining NER with Llama-3 (70B) via Groq, vector search, and graph traversal for complex multi-hop medical query resolution.

• Architected a production-ready e-commerce platform for electronics retail with dynamic inventory routing and secure payment integrations.
• Delivered a seamless shopping experience with responsive frontend, real-time stock management, and optimized checkout flow reducing cart abandonment.
• Implemented scalable backend architecture with secure authentication, order tracking, and admin dashboard for inventory analytics.