I design and ship production-grade AI SaaS platforms — handling everything from domain modeling and distributed system architecture to LLM integration, cloud infrastructure, and CI/CD pipelines.
Not prototypes. Not side projects. Working systems used by real people at scale.
Currently:
- Engineering Lead at Hackini — building the infrastructure layer for AI-powered hackathon and innovation platforms
- Software Engineer at Prospecter — building AI prospecting pipelines that automate lead discovery and enrichment
Scope of ownership: System architecture · API design · AI pipeline engineering · Cloud infrastructure · Team leadership · Product delivery
- Scaling Hackini to support concurrent multi-event programs with isolated tenant environments
- Building AI evaluation agents that replace manual hackathon judging with structured LLM-powered scoring
- Exploring agentic workflows using tool-calling and multi-step reasoning for automation
- Hardening DevSecOps practices: automated security scanning, dependency auditing, secrets rotation
Complexity is a liability. I favor systems that are easy to reason about over systems that are impressive to describe. Every abstraction needs to earn its place.
My approach in practice:
| Principle | How I apply it |
|---|---|
| Design for the read path | Schema, API contracts, and data access patterns are defined before any code is written |
| Fail fast, recover predictably | Explicit error boundaries, dead-letter queues, structured logging from day one |
| Treat security as a constraint, not a feature | RBAC, secret isolation, and input validation are baked into the architecture, not bolted on |
| Optimize for the 80% path | Profile before optimizing. Most bottlenecks aren't where you expect them |
| Own the deployment | If you can't deploy and monitor it, you don't fully understand it |
I design systems around domain boundaries, not technical layers. Each service owns its data, exposes well-defined contracts, and fails independently.
Patterns I apply consistently:
Domain-Driven Design
└─ Bounded contexts per business domain
└─ Aggregate roots to enforce invariants
└─ Domain events for cross-context communication
API Design
└─ Contract-first (OpenAPI / typed SDKs)
└─ Versioned endpoints, backward-compatible by default
└─ Rate limiting + request tracing at the gateway
Async & Real-Time
└─ Event queues for workloads that can tolerate latency
└─ WebSockets for live collaboration and real-time scoring
└─ Idempotent consumers to survive duplicate delivery
Data Architecture
└─ Read/write separation where query complexity demands it
└─ Optimistic locking for concurrent writes
└─ Structured audit logs for every state-changing operation
Security
└─ Multi-tenant data isolation at the query level
└─ JWT + OAuth2 with scoped claims
└─ Secrets managed via Azure Key Vault, never in env files
I integrate AI as a production system component, not a demo feature. That means reliability, cost awareness, and fallback handling — not just calling an API.
How I build AI systems:
LLM Integration
└─ OpenAI GPT-4 / Gemini for structured generation and evaluation
└─ Function calling / tool-use for agentic task execution
└─ Prompt versioning and regression testing on outputs
Retrieval & Search
└─ Embedding pipelines (text-embedding-3-small / ada-002)
└─ Vector similarity search for semantic retrieval (RAG)
└─ Hybrid search: BM25 + dense vectors for recall/precision balance
Pipeline Architecture
└─ Async AI jobs via worker queues (decoupled from request cycle)
└─ Streaming responses for real-time UX
└─ Output validation and structured parsing (Zod / JSON schema)
Reliability & Cost
└─ Token budget enforcement per pipeline stage
└─ Response caching for high-frequency identical queries
└─ Graceful degradation when LLM APIs are unavailable
└─ Latency monitoring and SLA alerting per AI endpoint
AI-powered platform for hackathons, innovation challenges, and startup programs. Built from scratch. Architected for multi-tenancy, real-time scale, and AI-native evaluation.
Hackathon infrastructure is a patchwork of spreadsheets, Google Forms, Discord servers, and manual judging. Organizers spend 70% of their time on logistics. Participants get zero structured feedback. Sponsors see no data.
A full-stack SaaS platform that owns the entire hackathon lifecycle — registration, team formation, project submission, AI evaluation, live leaderboards, and organizer analytics — in one system.
┌────────────────────────────────────────────────────────────────┐
│ Client Layer │
│ Next.js 14 (App Router, SSR) · TailwindCSS │
│ Real-time updates via WebSocket subscriptions │
├────────────────────────────────────────────────────────────────┤
│ API Gateway │
│ NestJS · REST + WebSocket · OpenAPI-documented │
│ JWT auth · Rate limiting · Request tracing │
├────────────────┬───────────────────┬───────────────────────────┤
│ Identity & │ AI Evaluation │ Event Engine │
│ Access (RBAC) │ Pipeline │ (submissions, scoring, │
│ Scoped JWT │ LLM + Embeddings│ timelines, webhooks) │
├────────────────┴───────────────────┴───────────────────────────┤
│ Worker Layer │
│ Async job queues · AI scoring jobs · Email/notification │
│ Dead-letter handling · Retry with exponential backoff │
├────────────────────────────────────────────────────────────────┤
│ Data Layer │
│ PostgreSQL (primary) · Redis (cache + pub/sub) │
│ Azure Blob Storage (file uploads) · Audit log table │
├────────────────────────────────────────────────────────────────┤
│ Infrastructure │
│ Azure App Service · Docker multi-stage · GitHub Actions │
│ Azure Key Vault · Environment isolation per tenant │
└────────────────────────────────────────────────────────────────┘
- Multi-tenant isolation — each organization operates in a scoped data context; no cross-tenant data leakage by design
- AI scoring engine — LLM-powered project evaluation structured against custom rubrics, with output validated via JSON schema before storage
- Real-time leaderboards — WebSocket-driven live score updates across thousands of concurrent participants
- Async job workers — AI evaluation jobs decoupled from the HTTP cycle with queue-based retry and DLQ handling
- Reduces hackathon logistics overhead by ~60% through automated workflows
- AI evaluation delivers structured feedback in seconds vs. hours of manual review
- Organizer dashboards surface insights that previously required post-event data exports
Stack: Next.js 14 · NestJS · PostgreSQL · Redis · Azure · Docker · GitHub Actions · OpenAI API
|
AI prospecting & lead intelligence Problem: B2B sales teams waste hours on manual lead research that produces inconsistent, low-quality data. Solution: AI-driven pipeline that discovers, enriches, and scores leads automatically — delivering structured prospect profiles ready for outreach. Highlights: LLM-powered enrichment · semantic similarity scoring · async processing pipeline
|
AI + VR public speaking trainer Problem: Public speaking anxiety is the #1 professional fear, yet there's no scalable, personalized practice environment. Solution: Immersive VR environments combined with real-time AI speech analysis — feedback on pacing, filler words, confidence markers. Highlights: real-time audio analysis · VR scene rendering · cross-platform (web + mobile)
|
|
AR + AI plant diagnostics Problem: Farmers and gardeners lack accessible tools to diagnose plant diseases, often acting too late. Solution: Point your camera, get an instant diagnosis. On-device ML for offline-capable detection — native on both Android and iOS. Highlights: on-device ML inference · AR overlay · cross-platform native (Compose + SwiftUI)
|
Multi-entity financial management Problem: IEEE student branches manage budgets across dozens of committees with no unified system — leading to audit gaps and reconciliation errors. Solution: Centralized multi-tenant accounting with role-based access, automated reporting, and full audit trail. Highlights: multi-entity ledger · role-scoped access · automated financial reports · containerized deployment
|
| Core |
|
| AI & Data |
|
| Mobile |
|
| Infrastructure |
|
I build systems that matter — and I'm selective about what I work on.
If you're building something ambitious at the intersection of AI, SaaS infrastructure, or developer tooling and need someone who can own architecture end-to-end, I want to hear about it.
Startup founder with a hard technical problem · CTO looking for a senior engineering partner · Team building something that needs to scale — let's talk.
