Founding AI Engineer at XXAVI AI Tech

I build AI products end to end, and the systems that keep them running.

Backend, apps and infrastructure — built so they stay reliable, safe and cheap to run. Right now that's Prajna, an AI storybook app for kids. Before that, two SaaS products I shipped solo, with paying users.

Selected work GitHub LinkedIn tpbh230@gmail.com
$0.48 → $0.11
AI cost per story, Prajna
86%
cheaper AI retries, Prajna
< 150 ms
p95 & p99 API latency, SaaS
40+
paying users, zero ad spend
01 — Selected work
01 / Founding AI Engineer, XXAVI AI Tech Live on Google Play

Prajna

An AI storybook app for kids. I built the whole platform: the generation engine, the backend, the Android app and the admin panel.

AI cost per illustrated story
$0.48 → $0.11
cheaper retries after a failure
86%
AI gateway overhead per call
< 70 ms
Read the case study Google Play prajnaai.org
02 / Solo founder and engineer Live

AttemptIQ

UPSC diagnostic analytics. Turns mock test results into what to fix next: decaying topics, trap patterns, real readiness.

paying users, zero paid marketing
15+
organic pageviews
12,000+
Case study attemptiq.com
03 / Solo founder and engineer Live

ScoreResume

AI resume scoring against a specific job description, with an async pipeline so LLM timeouts never fail a request.

paying users, zero paid marketing
25+
organic pageviews
18,000+
Case study scoreresume.com
02 — Experience
Jul 2026 — Now

Founding AI Engineer · XXAVI AI Tech

Sole engineer on Prajna: the AI generation engine, backend, Android app and admin panel.

  • Crash-resumable AI engine that saves every step, so a failure on image 6 of 7 re-runs 1 image, not the whole story. Retries 86% cheaper.
  • Transactional outbox between Postgres and SQS, with a reconciler. Zero lost or double-billed jobs.
  • AI gateway across 3 providers with circuit breakers per provider and task, a daily spend cap and < 70 ms overhead.
  • Cut AI cost 77% ($0.48 → $0.11 per story) by measuring cost per call. An LLM judge checks every image for 11% of its cost.
  • Fixed 4 billing bugs in a payments audit, including a webhook deadlock that hit 29 of 30 runs.
  • PII-safe logging for a children's app: a redactor 90× faster than pino's built-in that catches data leaking through error messages.
Jan 2024 — Jun 2026

Full Stack Developer · TheBrandwick

Led engineering for four international clients across EU, UK and US markets — sprint planning, client communication and delivery.

Show 4 highlights +
  • Booking and FAQ systems for quoka.co.uk, with 90+ Web Vitals.
  • Led architecture for VastuWisdom and mentored 2 engineers.
  • AlexandersFA: design to deployment, 95+ SEO, retention up 35%.
  • Frontend for AllUnity, contributing to a 45% rise in conversion.
2023

Full-stack program · Masai School

2017

B.E. Mechanical Engineering · Don Bosco Institute of Technology

03 — About

Mechanical engineer by training. Software builder by choice.

I'm Trilochanprasad BH, a Founding AI Engineer at XXAVI AI Tech in Bengaluru, India. I'm the sole engineer behind Prajna, an AI storybook app for kids on Google Play. Before that I built and ran two SaaS products on my own, AttemptIQ and ScoreResume, both with paying users.

I care about the unglamorous parts: jobs that survive crashes, money that can't be counted twice, and code that doesn't wake anyone up at 3 AM.

I work best owning a product end to end, from schema to deploy and billing to monitoring, and writing down what I learn along the way.

Languages
TypeScript, SQL
Backend
Node.js, Express, oRPC, Zod, Drizzle, Prisma
Data & queues
PostgreSQL, Redis / Valkey, SQS, BullMQ, pgvector
AI
OpenRouter, Gemini, OpenAI, LangChain, RAG, LLM-as-judge
Apps
React, Next.js, Astro, React Native, Expo
Infra
AWS (EC2, RDS, SQS, S3, CloudFront, SES), Cloudflare, Docker, Sentry, PostHog
04 — Writing
Oct 30, 2026 Ten System Design Ideas That Finally Clicked for Me I wrote myself a system design notebook with a few hundred chapters. These are the ten ideas that changed how I actually build software. 8 min Jul 1, 2025 On Building Solo: What Shipping Two SaaS Products Taught Me Lessons from independently building, launching, and growing two production SaaS platforms — AttemptIQ and ScoreResume — without a team, funding, or safety net. 5 min Jun 15, 2025 Building Async AI Pipelines with BullMQ and Redis How I built a production-grade async AI processing pipeline using BullMQ and Redis to handle LLM timeouts, retries, and dead-letter queues in ScoreResume. 4 min

All posts →

05 — Contact

Building something ambitious in AI or SaaS? I'd like to hear about it.

tpbh230@gmail.com
GitHub LinkedIn X / Twitter