I'm the person who takes an idea from design to deployed — writing the React components, wiring the Python agent, configuring the Docker stack, and shipping the binary.
Currently at Ofofo, I architected a cross-platform Electron app that runs AI-powered cloud compliance scans for enterprise clients across all major SCF frameworks.
I pick up whatever it takes — React one day, a Python MCP server the next, D3 force-directed graphs the day after. Lately I've been building AI-native chat interfaces where the assistant doesn't just talk back — it renders forms, score cards, drill-downs and audit exports inline as part of the conversation.
RAG pipelines with real user-facing UIs. MCP servers that resolve natural-language intent into structured tool calls. Multi-LLM routing, agentic workflows, local SLM fallback — and chat surfaces that render interactive components, not just text.
Frontend as the product, not an afterthought. SSR/SSG, route-level code-splitting, knowledge-graph visualisations with D3, dark-mode Electron desktop UIs. Complex flows that still feel light.
Python FastAPI agents, Node services, Docker on GCP, Stripe + Twilio integrations, electron-builder packaging with notarization. Idea → production binary, owned end-to-end.
Cloud compliance evidence collection across multiple frameworks was manual, slow, and error-prone for security teams.
Cross-platform desktop app — React + TypeScript renderer, Node.js main, bundled Python FastAPI agent, Steampipe + Powerpipe under the hood, electron-builder with notarization for macOS, Windows and Linux.
Supports the full SCF framework set (SOC 2, HIPAA, ISO 27001, NIST CSF, PCI DSS, GDPR, CIS, Essential 8, CIS CSC v8 and more) across GCP, Azure and AWS — three OSes from one codebase.
LLMs can talk; they can't query the underlying compliance state. The gap between chat and real cloud data is where most "AI for ops" tools fall apart.
Built an ontology-driven MCP (Model Context Protocol) server that maps natural-language prompts to the right framework controls and Steampipe queries, then feeds structured results back to the model for remediation suggestions. Local SLM fallback when offline.
Bridges chat UI to live cloud state via a tool-calling agent loop — answers grounded in real evidence, not vibes.
Plain-text chatbots can't actually do the work — they hand back walls of text. Real ops needs forms, controls and structured output rendered inline in the conversation.
An AI-native chat surface where the agent's structured tool outputs render as live components — forms, score cards, control drill-downs, snapshot diffs, audit-export buttons — all inline. Built the small UI library that maps typed agent tool calls to UI primitives.
The chat becomes an action surface, not a Q&A. Users complete real tasks without ever leaving the conversation thread.
Compliance teams sit on hundreds of internal docs — policies, runbooks, contracts. Grep is useless; semantic search is the answer.
End-to-end pipeline UI — upload with live progress, Flowise-backed chunking and embedding, semantic search front-end. Users query private compliance docs in plain English.
What used to be a 30-minute hunt through SharePoint became a 3-second answer with citations.
The Graphbit ontology had thousands of related entities. Lists and tables can't show the relationships — you need a graph you can grab, zoom, and follow.
D3.js force-directed graph backed by Neo4j queries via a Node service. Custom node typing, edge weights, hover-to-reveal metadata, click-to-pin, double-click to drill in.
Analysts navigate a graph of thousands of nodes in real time — the explore-loop that lists never gave them.
Single-provider LLM apps are brittle: rate limits, latency spikes, model drift. The product needs choice without rewriting the app per provider.
LiteLLM-powered widget that abstracts OpenAI, Anthropic, Gemini and local providers behind one chat surface. Per-prompt template overrides, session memory, automatic fallback when a provider rate-limits.
One widget, four providers, zero vendor lock-in. Outages in one provider auto-failover to the next without users noticing.
Bengaluru, India · 560076
Open to remote · hybrid · onsite