I build production systems — from AI safety infrastructure to full-scale SaaS.
3+ years shipping end-to-end: LLM guardrails at Tumeryk, a complete Business OS built solo for Nevioz, and freelance products live in production today.
Flowcraft — A Complete Business Operating System
Airtable + Jira + Zapier in one platform. Architected, built, and shipped single-handedly as a freelance engagement for Nevioz. Live with real users.
6 Views, One Data Model
Table, Kanban, List, Gallery, Timeline, and an Excel-like Grid with keyboard navigation (Tab/Shift+Tab/Enter), inline auto-save on blur, and frozen first column. Every view reads from the same normalized data model — switch instantly, nothing re-fetches.
How it's actually built.
Hybrid EAV + Denormalized Columns
Custom fields in field_values table. Default fields (title, description, due_date) as real columns. Reference fields in entity_references with nullable field_id for standalone links. Avoids per-customer migrations while keeping common queries fast and index-friendly.
Atomic Ref Numbers, No Locks
ref_counters table with UPDATE...RETURNING for atomic increment. Each org gets its own counter per entity type (FL1-3, CLT-2 format). Zero race conditions on concurrent item creation — pure Postgres atomicity.
STS Temp Credentials for Lambda
Backend assumes IAM role via STS AssumeRole per export request. Lambda receives scoped temp credentials — no long-lived keys stored anywhere. Compromise of one export session cannot escalate to full S3 access.
SSE Search, 4 Parallel Queries
p-limit concurrency control + Promise.allSettled. GIN indexes on all searchable columns via pg_trgm. Individual query time under 20ms. SEARCH_CONCURRENCY_LIMIT and SEARCH_SECTION_PREVIEW_LIMIT configurable via env.
Where I've shipped.
AI Safety Infrastructure for Production LLMs
- —Guardrail testing agent: auto-evaluates LLM policy compliance across risk categories, generates model trust score per response
- —Base policy enforcement layer: the core primitive gating every custom policy through organizational safety constraints
- —Automated LLM evaluation workflow: adversarial prompt sets validated through a judge-model pipeline scoring safety, accuracy, and guardrail compliance
- —PII validation rail: real-time detection preventing confidential data exposure inside live AI interactions
- —Observability dashboard: real-time guardrail metrics, evaluation outcomes, and system logs for safety and ops teams
Cloud Accounting SaaS — Led from Architecture to Production
Automated reconciliation cut clients' monthly accounting close cycles significantly.
- —Architected the full platform in Adonis.js + Vue.js with automated reconciliation and unified cross-department data access
- —Multi-firm invoicing engine: SGST, CGST, IGST + international VAT for clients across India and abroad
- —RBAC with dynamic per-module permissions; Razorpay subscription billing and plan management
- —GST auto-fetch and caching via AWS Lambda + third-party APIs — eliminated manual GSTIN lookups for accounting teams
- —Rewrote PDF/Excel generation pipelines; WhatsApp messenger integration for invoice delivery and broadcasts
- —CI/CD scripts for AWS EC2/ECS that standardized team deploys
Freelance — Shipped for Clients
End-to-end delivery: from first conversation to production deploy.
Flowcraft
Full Business OS for Nevioz — multi-tenant, multi-module, production-grade. Built solo.
GalaxIQ.ai
Marketing landing site for an AI company. Designed and built solo.
Shirisha Morem
Personal portfolio site. Designed and built solo.
Have a project in mind? Let's build it.
The toolkit.
GenAI & LLM
Backend
Frontend
Cloud & DevOps
On GitHub.
Loading GitHub activity…
Let's work together.
Open to senior full-time remote roles and freelance projects.