Mohit Kumar · AI consultant and Lead Software Engineer
I automate manual work with AI agents, then audit your AI systems for security holes and margin leaks. I have spent 8+ years building production backends in Python and Go, so the fixes ship as working code, not slide decks.
Each service has a fixed scope, a written deliverable and a way to verify the result.
Service 1
AI automation replaces repetitive manual work with LLM agents and event-driven pipelines that run without supervision.
You get: a working pipeline in your cloud, runbooks, and an evaluation set to track quality.
Service 2
An AI security audit finds how your LLM application can be abused, such as prompt injection, data leaks and over-permissioned agents, before an attacker does.
You get: a findings report ranked by severity, fix guidance, and a retest of the fixes.
Service 3
An AI margin audit shows what each AI request really costs and where your gross margin leaks, then cuts that cost without hurting quality.
You get: a cost breakdown, a ranked savings plan, and the changes implemented if you want them.
| Area | Finding | Severity | Fix |
|---|---|---|---|
| Security | Agent can call a write API with no approval step | High | Scoped credentials and human approval for writes |
| Security | Customer details written to prompt logs | High | Redact before logging, shorten retention |
| Margin | Same large model used for simple classification | Medium | Route easy requests to a smaller model |
| Margin | Long shared prefix recomputed on every call | Medium | Enable prefix caching |
| Automation | Retries re-run the whole pipeline | Low | Retry individual steps from a queue |
Every audit ends with a report in this format, plus a retest after your team applies the fixes.
A typical audit takes a few weeks, depending on how many systems are in scope.
We agree which AI systems, data and costs are in scope. I work from read-only access and your existing logs and billing.
I test the system, trace the cost per request, and rank every finding by impact and effort.
You get the report. I can implement the top fixes with your team and retest so you can see the improvement.
Results from systems I designed and shipped.
Recommendations from my LinkedIn profile and a client review from Upwork. Names are withheld to protect confidentiality. Read them all on LinkedIn.
Mohit has consistently demonstrated exceptional technical expertise, strong leadership abilities, and an unwavering commitment to excellence.
Mohit showcased exceptional skills in delivering scalable, production-grade applications.
Mohit is one of the few people who ramps up quickly on any problem statement. … I would recommend Mohit for anyone building a 0 to 1 product.
He played a pivotal role in debugging and resolving production issues, ensuring minimal downtime and optimal system performance.
Mohit is motivated, forward thinking and dedicated employee who has lots of knowledge of his field.
Mohit is a professional with a very high skillset. I HIGHLY Recommend working with him.
I am a Lead Software Engineer with 8+ years building distributed systems in Python and Go. Since 2024 I have focused on production AI: multi-agent systems, RAG pipelines, LLM observability and AI-assisted engineering for teams of engineers.
I started auditing AI systems because most failures I see in production are infrastructure and design problems, not model problems. Unsafe permissions and wasteful token spend are the two I see most.
Tell me what your AI system does and which service you need: automation, a security audit or a margin audit.