Skip to work

Production AI engineer · ex-JupiterOne · sold my AI SaaS

I take your AI app from demo to production.

Secured, cheaper to run, and trusted by your customers. I harden the app, lock down the AI, and cut your AI bill, so you can say yes to real users and enterprise deals.

Taking 2 new projects a month.

  • 40K+users on the AI SaaS I founded
  • Soldthat AI SaaS, Remote Leaps
  • ~45%AI cost cut on an enterprise assistant
  • OWASPLLM Top 10 hardened, in a SOC2 / GDPR product
  • 58open AI-security guides I maintain

How I work

Trust first. Then we go deep.

Before I touch your code, we build trust. Then I go deeper: into your business, your customers and how they feel when they use your AI.

  1. 01Build trustA small, clear first step. Honest findings, plain language, no surprises. You see exactly what I see.
  2. 02Understand your customersWhat they expect, what worries them, and where the AI lets them down. Their trust is what we are protecting.
  3. 03Fix what matters to themEvery fix is chosen for the people using your product, not just to pass a checklist.

Why AI apps break in production

Your demo works. Production is a different game.

  • 01SecurityPrompt injection, data leaks, weak auth and exposed keys. One incident can cost you your biggest customer.
  • 02CostAI bills that grow faster than revenue. Wasted calls, oversized prompts, no caching.
  • 03TrustHallucinated answers, silent failures, no monitoring. Customers stop relying on it.

I fix all three. It's what I did at JupiterOne and in the AI SaaS I built and sold.

Selected work

Proof, not promises.

Three AI products in front of real users, each hardened for security, cost and trust.

JupiterOne · 2025–2601
  • Security
  • Cost
  • Enterprise trust

Support cases solved 35% faster.

Juno

Enterprise AI security assistant

Hardened for SOC2 / GDPR enterprise customers, with ~45% less AI workload per question.

An AI assistant that answers big companies' questions about their own security, live, as they type. Answers stream in word by word with live tables of their systems.

  • Streaming answers + live data tables on AWS Bedrock
  • Hardened against the OWASP LLM Top 10 in a SOC2 / GDPR product
  • ~45% less AI workload per question with caching and summaries
  • React
  • TypeScript
  • AWS Bedrock
  • SSE
  • RAG
jupiterone.com
Remote Leaps · founded & sold02
  • Trust
  • Scale

40,000 users, then acquired.

Remote Leaps

AI résumé SaaS, built and acquired

AI built to never invent experience, trusted by 40K+ users in 25 countries, then acquired.

I started an AI résumé product, designed it, built it, and grew it to 40,000+ users in 25 countries before selling it. Its AI was built never to invent experience.

  • 40K+ users across 25 countries
  • Scoring model lifted interview rates 35%
  • Led a 4-person team: roadmap, architecture, hiring
  • Next.js
  • Fine-tuned scoring
  • Constrained decoding
See the timeline
Open source · 56 releases03
  • Reliability

An agent that fixes its own mistakes.

miii-cli

Open-source coding agent

An agent that tests its own work and rolls back anything that breaks.

Describe a change; it plans it, writes the code, runs your tests and keeps going until they pass, rolling back anything that breaks. Works with any model, even local ones.

  • 14+ AI providers, cloud or fully local
  • Runs tests until green, with automatic rollback
  • Checkpoints every turn; every risky step asks first
  • TypeScript
  • Node
  • Ink
  • MCP
Read the deep-dive

Services

How we work together

Start small with an audit. Fix what matters. Keep it safe.

  • 01

    Production Trust Audit

    ~1 week · Fixed price

    • Web app security review: auth, permissions, data exposure, secrets, rate limits
    • AI security review against the OWASP LLM Top 10 (prompt injection, data leakage, unsafe output)
    • AI cost analysis: caching, model choice, prompt size, wasted calls
    • Reliability & trust: hallucination guardrails, error handling, monitoring

    You getA scored report with your Trust Score, the top 10 fixes ranked by risk, and estimated monthly savings.

  • Most popular02

    Hardening Sprint

    2–4 weeks

    • I fix the issues found in the audit, highest risk first
    • AI guardrails, output validation and fallbacks
    • Caching and model routing to cut AI cost
    • Logging and monitoring so failures never go silent
    • Ready for enterprise security questionnaires
  • 03

    Production Partner (monthly)

    Ongoing, monthly

    • Monitoring of security, cost and AI quality
    • Monthly cost and risk report
    • Help with customer security questionnaires
    • Priority fixes and updates

Not sure where to start? The audit pays for itself in most cases through cost savings alone.

FAQ

Questions, answered.

Experience

Seven years, one exit.

From frontends to founding an AI company and selling it, then enterprise AI at a security company. I've led and mentored teams of 4 to 5 engineers. Today I help teams ship AI their customers can trust.

  1. Senior Full-Stack AI Engineer · India · remote

    • Helping SaaS products strengthen their cybersecurity.
    • Revamped the app with graded, measurable performance improvements.
    • Managed the multi-authentication overhaul and updated authorization for every sub-consumer of the app.

What I do

The model and the screen it lives on.

I'm Akshay, a production AI engineer based in India, working remotely with teams worldwide. Most AI projects stall between the demo and production. I take them across: the retrieval, the agent, the evals and security, and the interface people actually use.Off the keyboard, I train and share the honest side of getting stronger at @akshaay.lifts.

  • LLMs & RAGAssistants that answer from your own docs and data, with sources, not guesses.
  • AI AgentsMulti-step agents that plan, call tools and check their own work.
  • Evals & GuardrailsAutomatic quality checks on AI answers in production (RAGAS, Braintrust).
  • AI SecurityPrompt injection and data-leak reviews mapped to the OWASP LLM Top 10.
  • Streaming UIsReact and Next.js interfaces where answers stream in word by word, with live tables and no spinners.

Open source

Built in the open.

The AI-security playbook I maintain publicly. Free to use, read and fork.

  • mii-ai-security

    58 guides · OWASP / MITRE

    A security playbook for teams shipping LLM apps, mapped to OWASP LLM Top 10 and MITRE.

    Deep-dive →GitHub