Pranav Nagothu

Write-time flaky test prevention platform — catches flakiness before tests even run using AST static analysis and AI-generated fix patches.

• Go + Tree-sitter AST engine detects all 4 root causes of flakiness at write-time: async races, shared state, hardcoded timeouts, ordering bugs
• GitHub App webhook → AWS Lambda triggers analysis on every PR; LangChain + Claude API generates fix patches
• PyTorch binary classifier trained on 10K+ open-source tests predicts flakiness with explainability output
• Real-time D3.js heatmap with WebSocket updates — risk scores, trends, team analytics

Production-grade chaos engineering toolkit — HTTP reverse proxy with policy-driven fault injection, Redis pub/sub kill switch, and real-time blast-radius control.

• HTTP reverse proxy in Go injecting configurable latency, error rates, and network partitions via policy engine
• Redis pub/sub kill switch halts all active chaos experiments across distributed nodes in <100ms
• JWT/RBAC auth, audit logging, and blast-radius enforcement prevent accidental production harm
• React dashboard with live experiment metrics, history, and one-click rollback

Document processing and data transformation platform handling 5+ document types — 70% reduction in research time with 98% accuracy.

• Built a document processing and data transformation platform handling 5+ document types, reducing research time by 70% with 98% accuracy
• Designed a multi-step LLM pipeline with LangChain and persistent context, achieving <2s response time and reducing server costs by 60%
• Serverless Next.js + TypeScript frontend with Firebase backend enabling real-time document streaming and multi-user collaboration
• Deployed on AWS with auto-scaling — handles concurrent document processing across contract, filing, and brief document types

AI-powered codebase assistant combining RAG, AST analysis, Graph RAG, and agentic workflows — 85% reduction in manual review effort, 95%+ defect-detection accuracy across multi-language codebases.

• 5-stage retrieval pipeline: vector search (Chroma) → LLM file selection → ensemble (60/40 weighted) → Graph RAG context expansion → cross-encoder reranking (ms-marco-MiniLM) for precision-first results
• LangGraph agentic workflow with 6 tools (search_codebase, read_file, find_callers, find_callees, find_call_chain) enabling multi-hop reasoning over large codebases
• Tree-sitter AST structural chunking preserves semantic boundaries; Merkle tree incremental indexing re-processes only changed files across ZIP, GitHub, local, and web sources
• Multi-model support: Gemini Pro + Groq (LLaMA 3.3) with adaptive rate limiting and fallback chains; Streamlit + FastAPI + Next.js multi-interface frontend

Multi-agent financial analysis pipeline using LangChain and Gemini API — automates stock analysis, earnings sentiment scoring, and investment report generation with 85% reduction in manual research effort.

• Orchestrated a multi-agent pipeline using LangChain and Gemini API to automate stock analysis, earnings sentiment scoring, and investment report generation, reducing manual research effort by 85%
• Developed a FastAPI gateway with modular REST API endpoints and independent error handling per data source, serving 50+ concurrent users with 95% error recovery rate
• Containerized using Docker and automated CI/CD via GitHub Actions for reliable deployment and continuous delivery