# Skills and credentials: Surya Pugazhenthi

Source page: https://suryapugaz.com/license-center

## Credentials

- B.S. Computer Science, Baskin School of Engineering, University of California, Santa Cruz, 2026
- A.S. Computer Science, Diablo Valley College
- AlphaForge Cohort 3 Diploma, GTM Engineering (Clay Cohorts, completed September 21, 2026). Credential ID CC-DIP-ED973D12, verify at https://cohorts.clay.com/c/ed973d12-60ce-4e35-8c4d-b961ed4daeea

## Skills

### HTML / CSS / TypeScript / React fundamentals

The rookie license: shipping real front-end interfaces with React, TypeScript, and modern CSS.

- First laps in React (Silver): Built BenefitFinder's matching UI with Vite + React in 36 hours at CruzHacks — a React web app shipped under real time pressure.
- React on mobile (Silver): Built slugspace, a React Native app with swipe-based matching, saved searches, and student profiles — React component skills carried onto native.
- Typed, production UI (Gold): Designed, prototyped, and iterated the Calendarize UI in a Next.js/React codebase — TypeScript React components in a shipped product.

### Next.js apps, APIs, and deployment

Full-stack license: building and deploying complete Next.js products with real back-end integrations.

- Ship it live (Gold): Led Nodegent as Product Owner and lead developer — a Next.js + Convex + Clerk student dashboard integrating Canvas and Google Calendar APIs, running live at nodegent.app.
- Telemetry to launch (Gold): Built a serverless analytics pipeline for Calendarize (Next.js API route → PostHog) capturing every interaction, then launched the product on Product Hunt.
- 36-hour full build (Gold): Built BenefitFinder end-to-end in 36 hours — Vite + React front end, Firebase auth and real-time database, matching across 20+ eligibility criteria — winning Best Beginner Hack at CruzHacks.
- Engine swap under green (Gold): Migrated this live Next.js site from Vercel to Cloudflare Workers via OpenNext with zero downtime, rebuilding the edge layer — HSTS, a TLS 1.3 floor, rate limiting on the admin login, WAF rules, and the 308 redirect map across two zones — behind a verified DNS-level rollback and a CI gate that builds the real Worker.

### React Three Fiber and 3D web pipelines

Real-time 3D on the web with React Three Fiber and a Blender-to-glTF asset pipeline (International B).

- 3D on the grid (Gold): Built this site's 3D in React Three Fiber: the GT Café museum room, the garage's inspectable hero cars, and the boot-intro montage — lazy-loaded scenes, camera rigs, and data-bound spec sheets, live in production.
- Blender to glTF (Silver): Runs the Blender → meshopt-compressed glTF → gltfjsx pipeline behind the café and garage: lighting baked in Blender, exports validated, models landing as typed R3F components.

### Agent systems and LLM applications

International A: designing autonomous agents and LLM-backed applications that take real actions.

- Autonomous transaction (Gold): Built TripWeaver on the Claude Agent SDK — an autonomous business-travel agent that books trips end-to-end and pays with real rails — winning 2nd overall + the Stripe Prize at the Locus hackathon.
- Signals into scores (Silver): Built Credence, a B2B prospect-intelligence platform scoring trust and fit from live web signals via real-time scraping — 3rd place, Best Use of Apify API at the EF Marketing Agents hackathon.
- LLM actions, audited (Silver): Wired server-side Groq LLM actions into Nodegent with a full audit log and instant revoke on every AI action — LLM automation with a human-in-control safety layer.
- Semantic matching (Bronze): Used Gemini AI for semantic search across federal, state, and nonprofit programs in BenefitFinder, matching students to aid across 20+ eligibility criteria.

### Venture scouting and investment thinking

Sourcing and evaluating early-stage startups, and forming investment theses (Super License).

- Sourcing pipeline (Gold): As a Venture Scout at LvlUp Ventures, ran founder outreach as an instrumented pipeline — 20+ personalized outreaches into 10 founder calls, surfacing 3 companies for potential investment.
- Pre-seed sourcing (Silver): As a Venture Associate at 16VC (Jun – Oct 2026), sourced and evaluated pre-seed and seed opportunities across the SF Bay Area through demo days, founder outreach, and the builder community.
- Reps in the room (Bronze): Built investment judgment through Venture Starters — live startup pitching sessions and investor Q&A — reps that sharpen how startups get evaluated.
- Founder's seat (Silver): Pitched ClientSight — an eye-tracking attention-analytics startup — to a 1st-place, $500 win at the SlugAI Tech Startup Pitch Competition, seeing the venture case from the founder's side.

### Go-to-market systems: signal detection, segmentation, and outbound

The rally license: building go-to-market systems for markets the standard tools cannot see, where the buying signal is an absence and the identity layer has to be built underneath the platform.

- Recon in the dark (Gold): Ran a payment-processor detection scan across 2,932 YC companies, with cost and distribution predicted before the run and the result landing within five rows of the estimate — built on a four-state detection design so that "we looked and found nothing" never reads as "this input is unset".
- A clock for Clay (Gold): Built a Supabase snapshot layer that diffs each vendor fingerprint against its prior observation and posts classified change events back into the table, giving a platform that only knows what is true now the ability to answer what changed — with reason-code guards so a failed scan can never masquerade as a removal.
- Pace notes, hand-checked (Silver): Cut 909 scored accounts into segments and hand-checked 30 of them against the segment's own claim, then renamed the segment when the name asserted more than the scan had actually observed.
- Twelve signed sends (Silver): Sent 12 fully personalized, signed emails behind an Apps Script reply loop that finds a labelled thread, strips the quoted text, classifies it and lands it back in the table in under 60 seconds — with the manual reply join published as a known limitation rather than hidden.
