Founding Engineer Full-stack AI systems On-chain security ~6y in prod

I build what usually takes a team.

I own the entire arc — architecture, backend, frontend, AI pipelines, cloud infrastructure as code, and on-chain security. I scaled a consumer platform to 300K users as head of technology, ship production AI systems solo, and reported a valid finding in a $63K Code4rena audit.

SOC·01 · N1 REV 26.07 · 1 OWNER CORE · ARCH AI ENGINE SRAM SEC I/O · DATA PLL PHY
P/N AB-SOC-01 Verified
0KRegistered users
0Cities served
₹1Cr+Annual turnover
Code4renaValid finding · $63K
FIG.01 · DIE PLAN VIEW · SCALE 1:1
01  /  Ownership — transistors ▶ top metal Cross-section · 07 metal layers
DIE X-SEC. FULL-STACK OWNERSHIP — ONE ENGINEER REV 26.07 · BEOL · 7-METAL · 1:1
TRANSISTORS  ▶  TOP METAL 7 METAL LAYERS · 1 OWNER M1 LOCAL → M7 GLOBAL M7Product architecture M6Backend & APIs M5Frontend & UX M4AI pipelines M3Cloud infra · IaC M2Security & auth M1On-chain & audits SiSilicon · first principles P/N AB-STACK-07 PROCESS: PRODUCTION-GRADE

From the silicon up to the top-level bus — one person accountable for every metal layer. No hand-offs, no seams, no "that's another team's problem."

02  /  Featured build Production AI
Featured · 01

NextMantra

AI hiring platform
Solo · full-stack · production
Live · nextmantra.ai

An AI recruiter that runs the entire first round — live voice interviews, résumé-to-role scoring, and ranked shortlists — so hiring teams only ever meet the candidates worth meeting.

One engineer, the whole platform: real-time voice interviews on a low-latency speech pipeline, a multi-LLM scoring engine that grades each candidate against the role's own rubric, and the recruiter dashboards that turn it all into a defensible shortlist.

The deploy pipeline is trustless by construction: code ships AES-encrypted and tamper-evident, so a partner who alters what they shouldn't simply breaks their own service security that holds by structure, not by contract.

Spec sheetP/N NM-01
Solo1 engineer · full platform
Real-timeLive voice interviews
0-TrustAES-encrypted, tamper-evident deploys
End-to-endSource → interview → score
Stack Next.jsNestJSPostgreSQLVoice pipelineMulti-LLMAWS · IaC
03  /  Featured build Scale + unit economics
Featured · 02

Urban Culture

At-home salon marketplace
Head of Technology · 2020–2025 · 6 cities
Live · urbanculture.me

Built and ran the entire technology for an at-home salon marketplace — four apps, custom infrastructure, and the systems that made the unit economics actually work.

A four-app ecosystem — customer app, partner app, admin console, and web — on a Flutter + Node + Firebase stack I architected and operated for four and a half years across six cities.

When Google's Distance-Matrix and geocoding bills threatened the margins, I replaced them with a custom geolocation service −80% cost — accuracy held.

Then the systems that moved the numbers: a targeted coupon engine +16% cart, a partner loan-management system +30% onboarding, and a Firestore → BigQuery pipeline feeding every decision.

Measured resultsP/N UC-02 · as shipped
0KRegistered users
0Cities served
₹1Cr+Annual turnover
0%Geolocation cost
+0%Avg cart · coupons
+0%Onboarding · loans
Stack FlutterNode.jsFirebaseCloud FunctionsBigQueryLooker
04  /  Component index Selected · 06 systems
P/N
KAF·1

Turns chaotic, spreadsheet-and-email customer onboarding into a repeatable, SLA-tracked pipeline — so customer-success teams ship predictable go-lives instead of chasing threads. Create a company, assign a workflow, assign an agent; managers watch bottlenecks before they become delays.

3-step activationSLA-tracked stagesmulti-tenantimmutable audit trail
Enterprise SaaS
Customer success
P/N
BNB·2

A yoga & meditation studio taken end-to-end — public site, class / workshop / retreat booking, and ₹ payments — fronted by a free wellness-assessment quiz that converts cold visitors into booked sessions and first-party data the studio actually owns.

3 apps · site · admin · APIassessment funnel → bookingRazorpay ₹ paymentsdesign + product + infra
Wellness · client
End-to-end
P/N
TE·3
Testing Engine In development

An AI precision testing engine for competitive-exam prep: it diagnoses exactly what's holding each learner back, maps the gap against their target-exam blueprint, and drills precisely that — to move the one number that matters, passing probability. Sold direct to aspirants and to the coaching institutes that train them.

adaptive diagnosticspassing-probability modelB2C + B2B2Cexam-day simulator
EdTech
Building now
P/N
AV·4

An always-on sourcing analyst: it reads inbound résumés straight from Gmail, scores them, and surfaces the strongest — unattended. Engineered around Firestore's hard query limits with cursor pagination and a bounded concurrency pool so it never trips a quota.

Gmail → résumé → scoreunattendedconcurrency-bounded
Internal tool
Automation
P/N
SEC·5
Blockchain & Security

Adversarial security work: a valid finding in a $63K Code4rena audit, formal verification with Halmos & Certora, low-level EVM (Huff / Yul), and Sui / Move — the same threat-model-everything discipline I bring to every system I build.

$63K audit · valid findingformal proofsEVM · Huff / YulSui / Move
Security
Adversarial
P/N
AI·6
Brahmo

Grades long-form exam answers at scale: an event-driven, multi-LLM pipeline — Gemini OCR → semantic matching → Claude scoring on a weighted rubric — with human-expert override where it counts.

multi-LLMOCR → match → scorehuman-in-the-loop
AI pipeline
Mains evaluation
05  /  Engineering notes How I build

I don't accept "should work." I trace every path to bedrock.

Note 01

Trace to bedrock

Decompose a problem to its smallest provable truth, then build up. Fix the root cause — never patch the symptom.

Note 02

Fail safe, not silent

Design so that when an assumption breaks, the system stops — it doesn't quietly lie. Bad behaviour should be self-defeating.

Note 03

Structure over trust

Trust isn't a security feature. Guarantees come from how a system is built, not from how carefully it's operated.