Open-source multi-agent orchestration system, a production-grade alternative to Manus. Co-architected the core framework for coordinating autonomous agents at scale.
Open-source, production-grade multi-agent orchestration
The Work
Open-source autonomous browser agent. Multi-agent architecture with context engineering, LLM orchestration, and DOM manipulation. Building agents that navigate the web the way humans do, except they don't get distracted.
91.2% on WebVoyager benchmark, outperforming OpenAI Operator (87%)
Knowledge Graph Memory
Neo4j-based long-term memory for agents. Custom semantic ingestion and retrieval that goes beyond typical RAG window limitations. Giving agents actual memory, not just context window hacks.
Persistent agent memory across sessions via knowledge graph retrieval
LDRM
Long Document Reasoning Model. A document processing pipeline built for financial analysis use cases. Handles large-scale document ingestion with high accuracy across complex, multi-page financial documents.
1000+ documents processed at 95%+ accuracy
Client Delivery
Led 4 end-to-end client projects spanning grants automation, chemistry, marketing, stock trading, and medical domains. The same agent architecture principles applied across wildly different domains.
4 production deployments across 5 distinct industries
The Research
Shipping production systems showed me where models actually fail. Research is how I try to understand why.
A new generative network using self-attention + conditional generation + label smoothing for satellite image synthesis that outperforms existing methods.
Springer Nature, Intl. Journal of Information Technology, 2025
Compared 4 GAN variants for satellite image augmentation. CDCGAN achieved the highest classification accuracy.
IEEE ICESC 2024, 2024· Cited by 4
Explored how LLMs and vision-language models reshape marketing content creation workflows.
Welingkar 5th Intl. Conf. on Sustainability, Leadership & Technology, 2024
What I Believe
Building production AI systems taught me that the gap between demos and deployment is where most agents die. The demo works on a happy path. The real world has no happy path.
I think the next breakthrough won't come from bigger models. It'll come from better architectures for learning, memory, and reasoning. Scale is necessary but not sufficient.
My background is in physics, chemistry, and mathematics. I believe the deepest problems in AI will eventually be solved by people who understand the fundamentals beneath the code. If you're working on agents, reasoning, or anything that makes AI less brittle, I'd love to talk.
Resume
Experience, education, and the full stack.