Software Engineer and AI Systems Builder focused on reliable, evidence-first AI systems.

I design products where AI decisions need to be useful, inspectable, and defensible. My work focuses on multi-agent coordination, grounded reasoning, durable decision records, and the infrastructure required to move agent systems from demos into production.

My product thesis is simple: capable AI is not enough. High-consequence workflows also need evidence, disagreement, provenance, and accountability.

I develop open-source and private projects around that thesis. The first public project is Caucus.

Caucus is an open-source decision layer for multi-agent systems. It grounds deliberation in live MCP evidence, preserves dissent, and produces a hash-chained decision record that can be independently verified.

• Provider-agnostic agent backends
• Evidence collection through the Model Context Protocol
• Adversarial review and confidence-aware synthesis
• Tamper-evident logs with an open record specification
• Prompt-injection boundaries designed into the orchestration layer

git clone https://github.com/srinath-jukanti/caucus.git
cd caucus
uv sync
uv run caucus deliberate "Adopt library X for feature Y?" --evidence evidence.json

• Agent infrastructure and multi-agent systems
• AI reliability, evaluation, and safety
• Evidence-grounded decision support
• Developer tools and applied AI products
• Auditable systems for regulated or high-consequence work

I am interested in senior software engineering and AI engineering opportunities, as well as conversations with researchers and engineering teams working on reliable agent infrastructure and applied AI. More technical work will be published here as it matures.