I build AI platform infrastructure, RAG pipelines, backend architecture, and production SaaS that stays secure when it ships.
8 months at Agentnomics.ai. 50+ PRs merged to production. This is the story of how that happened - scroll through it.
Nov – Dec 2025 · Chapter 1
No sandbox ramp-up. First week in production, I started finding what shouldn't be there.
- Hardcoded API keys in deploy.sh - found in week one, fixed immediately
- Auth bypass on Salesforce routes
- Plaintext credentials in a connections API GET response
3 critical vulnerabilities caught and fixed in the first two months.
Jan 2026 · Chapter 2
One 6,200-line server.py was the entire backend. Every PR was a merge-conflict risk. Parallel development was blocked.
6,200 lines
Activated 16 Flask blueprints, fixed relative imports across 11 files, reduced the entry point to 109 lines. Zero logic changes, zero regressions. Every backend PR since is scoped to a single blueprint - this was the unlock that made everything else possible.
Python · Flask · SQLAlchemy · PostgreSQL · Alembic
View Case StudyFeb 2026 · Chapter 3
The restaurant bot couldn't answer from real menus. So I built the RAG pipeline that let it - end to end, from scratch.
pgvector similarity search at chat time - 12 chunks, 0.25 threshold - with graceful fallback to raw text if pgvector is unavailable. Deployed and verified before a live client visit.
Python · Flask · pgvector · PostgreSQL · OpenAI
View Case StudyMar 2026 · Chapter 4
Twilio rejected A2P 10DLC registration. Twice. The voice product was completely blocked - so I found the way around.
Self-initiated a 6-vendor CPaaS evaluation across cost, compliance, and API compatibility. Recommended SignalWire as a drop-in Compatibility API replacement. The CEO funded the account within hours - the evaluation cost ~3 hours; the alternative was weeks of compliance limbo.
SignalWire · Python · Flask · Twilio Compatibility API
View Case StudyApr – Jun 2026 · Chapter 5
Then it was time to scale - and keep shipping.
Phase 5 frontend migration
Feature-slice architecture, 30+ files, zero regressions.
Stripe billing end-to-end
12 products, 4 categories, webhooks, checkout → trial activation confirmed.
POS integrations ×5
Toast, Clover, OpenTable, Resy, SevenRooms - credential masking, test-connection patterns.
10-gap client readiness audit
Self-initiated before a live customer visit. Fixed all 10 in a single branch.
Jul 2026 · Now
Voice bots with mid-call human handoff. LangGraph multi-agent workflows. LiteLLM. 50+ PRs in. Still shipping.
From security fixes to architecture decisions in 8 months. The next chapter is a full-time role at an AI startup or mid-stage SaaS company - backend, full-stack, or AI infrastructure.
Epilogue · Off the clock
What I build when nobody's asking.
AI Triage System
AI agent that triages inbound return/refund emails into tickets, cites store policy, and safely auto-executes low-risk Stripe refunds - escalating edge cases to CSRs. Fail-safe by design: the agent only acts where the blast radius is small.
Next.js · Inngest · Stripe · Neon · Drizzle · Clerk
PromptShrink
[AI TOOLING]Reduce token usage while preserving meaning. Compresses prompts intelligently to cut API costs without degrading output quality.
Next.js · OpenAI · Anthropic SDK
PRChangelog
[AI / SAAS]AI-generated PR changelogs - connect your repo, get a polished changelog on every merge. Automates the part of shipping nobody wants to write.
Next.js · Anthropic SDK · Firebase · Stripe
SaaScribe AI
[AI / SAAS]Transform PDFs into intelligent conversations - upload documents, embed them, and chat with your content.
Next.js · LangChain · Pinecone · Neon
CallSage
[FULL-STACK / AI]AI-powered video conferencing with custom AI agents - real-time video, auth, async agent jobs, and database built end-to-end.
Next.js · Stream · OpenAI · Inngest
AICortex
[AI INFRA]Reduce LLM spend by 20–40% without degrading output quality - multi-provider model routing across Anthropic, OpenAI, and Mistral.
Next.js · Anthropic SDK · Mistral
Filters, more builds, and everything in between.
All projectsThe next chapter
Let's build the next chapter
Looking for a full-time role at an AI startup or mid-stage SaaS company. If you're building something that ships to production and needs an engineer who cares about security and reliability - let's talk.