Building AI agents & knowledge systems.
Hello! I'm Mahesh Jaganiya, a senior full-stack engineer based in India. I've spent 6+ years shipping AI-agent and knowledge-graph systems: agent runtimes, semantic graphs, and the tooling that makes them usable.
Current Focus
I'm a Senior Full Stack Developer at MithyaLabs, building two platforms: an unannounced multi-agent software development platform I architected solo, and Comake, a semantic-graph substrate where I build the authoring tooling.
What I'm building
Comake
A substrate where an organization authors its domain (schemas, mappings, capabilities, policies, even UI) as a live semantic graph the engine interprets at runtime. The data is the source of truth: edit the graph and behavior changes, no codegen, no deploy. Integrations route through one canonical model, N+M instead of point-to-point glue that multiplies. My piece is the authoring tooling: the visual mapping workbench, the Electron Studio IDE that runs a live engine in-process so specs validate and preview as you author, and an engine-native agent runtime where the agent's tools and policies are themselves graph data.
Tech Stack
My Journey
I started with simple web apps and ended up in data-heavy systems: search, graphs, event-driven backends. The last few years have been about AI agents, and specifically the unglamorous parts that make them dependable: orchestration that survives crashes, approval layers, sandboxing, and the tooling for authoring what an agent is allowed to do.
Graph & Ontology
The semantic substrate behind Comake and SKL: ontology modeling, RDF, live graph queries.
AI & Agents
Multi-agent orchestration, approval architecture, and RAG: the core of the agent platform.
Backend Systems
Event-driven services with durable state: outbox, idempotency, one event per transition.
Apps
Web with React and Next.js, desktop with Tauri and Electron, plus iOS and Android.
Open Source & Community
When I'm not building for work, I'm usually hacking on open-source tools. Some recent ones:
- OpenAI Codex MCP support: contributed first-class Model Context Protocol support to the Codex CLI agent.
- JSON2Cypher: converts JSON data directly into Neo4J Cypher queries.
- sitemap-sage: a scraper that works from sitemaps, built on Jina AI.
- cli-upkaran: an extensible CLI toolkit.