Harshit,Duggal
Agentic engineer I build infrastructure for AI agents
Building Opal, an AI CRM for job hunting and matching — 28 users, 9,912 jobs surfaced, 29 resumes parsed, 1,258 matches.
Also building Sapphire, an AI-native B2B customer acquisition system for market research, ICP development, prospecting, enrichment, and autonomous marketing.
I build AI products from the ground up, from agent infrastructure to the software people use. Here's some of the work
Sapphire
AI-native B2B customer acquisition through research, prospecting, enrichment, and autonomous marketing.
Prism
Prism turns a natural query into quality research on companies and prospects with real intent.
PainiteHQ
AI abuse prevention infrastructure that controls agent spend, actions, and runtime.
AI Agent Budget
AI agent budget enforcement for limiting model cost, tokens, steps, and runtime.
AgentX CLI
Agentic CLI and orchestration infrastructure for running 50–100 agents in parallel.
The stack I work with
The technologies I use across AI agents, data, infrastructure, and production systems every day.
TypeScript
Next.js
Drizzle / Neon
AWS
OpenAI Agents SDK
LangGraph
Agent orchestration / tool calling
LLM evaluation + observability
Inngest
Stripe
Cloudflare
Gemini Agents Kit
BetterAuth
I build for the product, not the resume
I care about what the product feels like to use: clear interfaces, fast feedback, and systems that solve the problem without unnecessary complexity.
I build AI products end to end, working across architecture, agent orchestration, infrastructure, and UI. On Sapphire, that means a B2B acquisition platform, multi-agent workflows, enrichment infrastructure, and 40+ connected tools.
I keep the architecture simple until the product needs more. When something breaks, I trace it back, fix the smallest thing that matters, and check that it stays fixed.
AI writes code I still decide
AI can write code fast and handle parts of the engineering work. It can produce something competent, correct on paper, and still be completely forgettable.
The harder part is deciding what to build, what to leave out, and whether the result actually deserves to exist. I work with multiple coding agents, review what they produce, and own the architecture, tradeoffs, implementation, and result.
Building products people genuinely care to use, and knowing what is worth building.
Overengineering, unnecessary features, and code that exists only to look impressive.
I build systems that let AI do real work
Agentic systems need more than a model. I design the orchestration, tools, context, and execution around them so multiple agents can work together without turning the system into disconnected workflows.
I build AI systems that ship, stay reliable, and solve real problems.
I work across agents, models, APIs, databases, and product interfaces.
I take features from a rough idea through architecture, deployment, and iteration.
I use AI tools to move faster, then make the calls they cannot make for me.