Kesav Patneedi

An AI system that autonomously finds and analyzes financial documents to answer company-specific queries.

• Built a supervisor–subagent architecture to autonomously discover, download and index SEC filings, annual reports and financial documents.
• Applied Astute RAG for iterative, source-aware knowledge consolidation, resolving conflicts across multiple retrieved documents for reliable responses.
• Implemented multi-hop query decomposition and financial jargon expansion to improve retrieval accuracy.

• Multi-Agent
• Astute RAG
• LangGraph
• Vector Search

An autonomous trading system that makes portfolio decisions using collaborative AI agents.

• Built a real-time data pipeline ingesting market, news and social signals using Redis Streams and MongoDB.
• Developed a PPO-based reinforcement learning agent using FinRL for autonomous portfolio allocation.
• Designed a multi-agent framework combining market, news, social and SEC signals for collaborative decisions.
• Implemented automated trade execution and portfolio tracking using the Alpaca API.

• Reinforcement Learning
• FinRL
• Redis Streams
• MongoDB
• Alpaca

Block ciphers implemented as neural networks, secured against key retrieval attacks.

• Implemented PRESENT-80 and AES block ciphers entirely as PyTorch neural networks with fixed deterministic weights, encoding all cryptographic operations as custom NN layers.
• Designed XORNet (ReLU-based), SBoxLayer (corner-function detection) and PermutationLayer (sparse matrix) as non-trainable PyTorch modules replicating exact cipher operations.
• Hardened the implementations against key retrieval attacks, preserving cryptographic security properties within the neural network framework.
• Verified correctness against all official test vectors with full pytest coverage across S-box, P-layer, key schedule and round key generation.

• PyTorch
• Cryptography
• PRESENT-80
• AES
• pytest