whoami
I'm Sudharshan. I build and market products, and take them from zero to one.
Across 4 startups over 8 years, all with the same founder, my hero story is SaaSworthy: we bootstrapped it to ~1M users and ~$6M ARR with a 9-person team, and a global top-10 spot in software research.
That work put me on India Quotient's 30 Under 30.
More recently, I built my own AI tools to run marketing. At way2top, a career-coaching startup, I ran the paid ads leveraging Claude Code and turned ~1,000 trials into 30+ paying executive-coaching clients.
now
Building Curcuma, launching shortly. Also open to ambitious product and growth roles, remote or relocating.
work
ls ~/work
-
curcuma
built solo · launching now
Email intelligence for coding agents. A free MCP backed by 20,000+ real production emails from 4,000+ brands: research any brand, benchmark your emails, write from examples that actually got sent.
-
way2top
eir, growth · 2025–26
AI career coaching. I built the AI-native ads stack (multi-agent ad strategist, server-side attribution) and turned ~1,000 trials into 30+ paying executive-coaching clients.
-
theCRUX
eir, go-to-market · 2025–26
AI enablement studio, and its product-market-fit bet. Led go-to-market and taught senior leaders to ship real products with Claude Code.
-
Intoglo
senior pm · 2023–25
Cross-border freight. Shipped the company’s first shipment-tracking product (NPS 80, 63 clients) and led AI document automation that cut manual data entry 80% at 97% accuracy.
-
HS Code Scanner
senior pm · intoglo
“Shazam for HS codes”: snap a product photo and get the US HS code plus tariffs in an instant. Built for customs brokers and exporters.
-
SaaSworthy
founding pm & marketer · 2018–23
Software research platform. Grew it from zero to ~1M monthly users and $6M ARR with a 9-person team, reaching the world top-10 in its category.
ai engineering
I spend most of my working day inside Claude Code. It runs my ad campaigns and my research, and it built this site. Here is what that looks like in practice; every line is backed by shipped code or a number:
- evals & llm-as-judge
- blind 3-judge A/B runs (Claude, GPT, Gemini), pre-registered prompt batteries, and a search-relevance eval scoring precision@5 = 0.95, all in Curcuma’s eval suite
- grounding & anti-hallucination
- “cited, not invented”: Curcuma forces agents to ground email copy in real references, with a harness that catches fabricated stats in transcripts
- guardrails & abuse control
- prompt-injection screening and PII masking over a 20k-email untrusted corpus, per-token quotas, and corpus-exfiltration canaries
- mcp & tool development
- Curcuma is a hosted MCP server (6 tools) with OAuth discovery, install-token minting, and one-paste installs for Claude Code, Cursor, and Codex
- retrieval (rag)
- BM25 full-text search fused with precomputed structural design signals over ~20k emails; picked keyword + structured retrieval over a vector DB on purpose
- structured extraction
- ~19.8k emails classified into schema-enforced JSON with a closed tag vocabulary, QA’d by two-model disagreement against blind human hand-labels
- agentic workflows
- a 3-persona ad-strategist system (conservative / aggressive / textbook) fanned out over a Meta Ads MCP to design way2top campaigns
- experimentation & attribution
- server-side Meta CAPI + PostHog + GA4; 6 documented ad test cycles, including catching Audience Network faking 100% of an ad set’s leads
career
- way2topEIR: co-created a coaching startup and acquired customers fast2025–26
- theCRUXEIR: led go-to-market; taught leaders to build with AI2025–26
- IntogloSenior PM: built the product for a shipping startup2023–25
- SaaSworthyFounding PM and marketer: grew it to ~1M users and ~$6M ARR2018–23
writing
How I think about products and growth: