Arnav Panigrahi
Software Developer, AI/ML Enthusiast
About Me
Software developer with an MS in Computer Science from UC Riverside, specializing in backend systems, distributed infrastructure, and AI/ML. Over 2.5 years of professional experience shipping production-grade software.
At AllCheer, I built a HIPAA-compliant AI system that reduced processing time by 60% and saved $20K annually. I'm focused on scalable systems, ethical AI, and open-source development.
My Skills
Languages
- Python
- JavaScript
- TypeScript
AI & Machine Learning
- LangChain
- LangGraph
- LLMs
- RAG
- PyTorch
Backend
- FastAPI
- Flask
- Node.js
- Express.js
- REST APIs
Frontend
- React
- Astro
- Angular
- Tailwind CSS
Cloud & DevOps
- GCP
- Docker
- CI/CD
- Git
Databases & Vector Stores
- PostgreSQL
- MongoDB
- Redis
- Pinecone
My Projects
bedtime.ai – Multi-Modal AI Storytelling Platform
A multi-modal AI pipeline that transforms children's drawings into personalized stories narrated in a cloned parent's voice. Integrates a fine-tuned phi-3 LLM, EfficientNet for image analysis, and coqui-ai for voice synthesis. Containerized with Docker; delivers complete stories in under 30 seconds.
Abliteration: LLM Uncensoring via Weight Modification
Techniques for removing censorship from LiquidAI/LFM2.5-1.2B-Instruct by identifying and orthogonalizing refusal directions in residual stream activations. Includes inference-time intervention and direct weight modification approaches that successfully uncensor content while preserving model capabilities.
Edge-Native Real-Time Web Platform
A real-time cursor sharing platform with YoRHa-inspired aesthetics from NieR: Automata. Uses Cloudflare Durable Objects to coordinate WebSocket connections across edge nodes, enabling visitors to see each other's cursors live. Built with SvelteKit and deployed entirely on Cloudflare's edge network.
Rethink BH Automation – FastAPI + Cloud Run Backend
A production FastAPI service syncing real-time appointment and authorization data from Rethink BH to Supabase via secure user login emulation. Achieves 90% cost savings and reduces sync latency from 24-48h to real-time; deployed with Docker, Secret Manager, and full observability on Google Cloud Platform.
RAG for Document Intelligence
A high-throughput RAG system for semantic search over PDF datasets. Uses Pinecone for vector storage and Groq for accelerated inference, achieving 35% improved search relevance over keyword-based methods with sub-second response times. Custom chunking strategies optimize retrieval accuracy.
Enterprise ROI System with AI Agent
A fully automated, AI-driven system that re-engineers manual Release of Information (ROI) workflows. The AI agent manages the entire document lifecycle with enterprise-grade security, delivering $20K annual savings and 60% reduction in processing time while adhering to HIPAA compliance.