citizenhicks
indie ml engineer
About Me
indie ml engineer and ai enthusiast specializing in multi-agent systems and autonomous ai architectures. building the next generation of multi agent applications that enable complex startegic modeling, planning, and tool use. passionate about creating intelligent systems that can decompose problems, coordinate across specialized agents, and deliver sophisticated solutions through orchestrated workflows. focused on pushing the boundaries of what agents can achieve in real-world applications especially in model development and validation in the fincancial services sector.
Projects
ThinkingSand Multi-Agent Platform CLOSED BETA
Comprehensive multi-agent AI platform featuring specialized agent systems with MCP (Model Context Protocol) support. Built on LangGraph and powered by multiple LLM providers, this platform orchestrates autonomous agents for deep research, financial analysis, code inspection, and actuarial modeling with advanced tool integration and real-time collaboration.
- Open Deep Research: PhD-level research agent with MCP integration, achieving #6 ranking on Deep Research Bench leaderboard ( this is an fork of the LangChain's Open Deep Research project)
- Artemis: Advanced actuarial analysis system with 6 specialized agents for insurance product analysis and pricing discovery
- ThinkingSand: A general purpose agent based on GPT-OSS-120b
- MCP Servers: Custom MCP implementations for real-time data integratiopn for the specialized agents to use
- Advanced agent orchestration with tool use, strategic planning, and autonomous workflow execution
Production-ready multi-agent platform demonstrating state-of-the-art agent coordination, MCP protocol integration, and scalable autonomous system architecture across multiple specialized domains.
pegasus
Multi-agent financial analysis platform featuring and ai agent that processes complex financial documents with tool use. Pegasus implements sophisticated agent workflows where specialized agent handle document ingestion, analysis, and report generation through task decomposition and tool use.
- Agentic system with a single ai for document processing, analysis, and synthesis
- Agent-driven PDF parsing with context-aware chunking strategies and memory systems
- Tool-using agent with access to multimodal capabilities for comprehensive document understanding
Entropix
Closely following and contributing to the Entropix project - an innovative approach to adaptive sampling and dynamic inference strategies for language models, enabling more intelligent and context-aware generation through entropy-based decision making.
GRAPES: Graph-based Reasoning and Planning with Ensemble Systems
An advanced multi-agent reasoning framework that enhances language model capabilities through Monte Carlo Tree Search. Implements strategic planning and ensemble decision-making for complex problem-solving tasks.
- Tech Stack: Python, asyncio, Pydantic, Agent Orchestration
- Achievement: Improved accuracy from 62.50% to 78.12% on complex reasoning tasks through dual-agent collaboration
- Features: Tree-based search strategies, ensemble voting, adaptive agent selection
TD3-HER for Robotic Control
Implementation of autonomous robotic agents using Twin Delayed Deep Deterministic Policy Gradient with Hindsight Experience Replay. This project demonstrates agent-based learning in sparse reward environments, where robotic agents must discover successful strategies through autonomous exploration and goal-conditioned learning.
- Tech Stack: Python, TensorFlow, Gymnasium, Panda-Gym, RL
- Features: Autonomous goal discovery, experience replay for agent learning, twin-agent architecture for stable training
- Applications: Robotic manipulation tasks with autonomous agent control
CyberSweat (also known as workout.AI)
One of the first applications to leverage generative AI for personalized fitness planning. CyberSweat uses GPT-3.5 as a backend that understands user goals, preferences, and constraints to generate fully customized workout plans. The app adapts recommendations based on user feedbacks.
- Tech Stack: Swift, SwiftUI, Firebase, OpenAI API
- Features: Pioneering use of ai for dynamic workout generation and personalization
- Impact: Early demonstration of practical applications in health and fitness
closed sourcecode