– Education
Diploma in Computer Science and Engineering
Thiagarajar Polytechnic College
Founder profile
Vijay K
Founder, MiniAI Labs
- AI Automation
- AI/ML
- DevOps Engineer
Agents, ML pipelines and cloud systems in production. I scope, build and hand over every MiniAI Labs project myself, from Bengaluru, India.
- Based in
- Bengaluru, India
- Experience
- 2+ years of DevOps and AI engineering
- Serves
- India, United States, United Kingdom, United Arab Emirates
- Credential
- AWS Certified Solutions Architect – Associate
Profile
An engineer who came to AI through operations
AI automation, AI/ML and DevOps engineer building LLM agents, RAG pipelines and n8n workflows on OpenAI, Claude and Gemini, MLOps pipelines on SageMaker and MLflow, and the AWS, Docker and CI/CD stack they run on — plus technical SEO and answer-engine optimisation (AEO) for client sites.
The work started in infrastructure. From May 2024 I looked after Linux servers and Kubernetes workloads, wrote modular Terraform for EC2, S3, RDS and VPCs, automated jobs with Lambda and EventBridge, and set up the Prometheus and Grafana alerting that tells a team when something breaks.
In May 2025 I was promoted into MLOps: versioned training, testing and deployment pipelines on SageMaker and MLflow, CI/CD for EKS, and AI-assisted monitoring triage on Ollama and AWS Bedrock. From November 2025 the work was voice agents, retrieval pipelines with cited sources and multi-agent automations across CRMs, WhatsApp and internal tools, for several clients of one agency.
That order shapes how I build. The first working version of a system already carries the parts that keep it running: retries on external calls, a trace of each run, prompts under version control and rolling releases designed to avoid downtime. MiniAI Labs offers that same engineering directly.
Architecture
How my agents are wired
The system I build most often, in the order a request moves through it. Four steps carry the request; four supporting pieces keep the agent grounded, capable, consistent and safe to ship. Select any piece to highlight it in the diagram.
Diagram of how Vijay K wires his AI agents: a request flows from trigger (Webhook, call, WhatsApp), then orchestrate (n8n, FastAPI, async, retries), then AI agent (Claude, GPT, Gemini, Ollama), then actions (CRM, APIs, WhatsApp). The AI agent draws on knowledge (pgvector RAG), tools (Functions, MCP), memory (Redis sessions) and evaluate and ship (RAGAS evals, tracing, Docker, CI/CD, AWS EC2).
The request path
What the agent draws on
Why each piece is there
The request path
- Trigger
- A run starts from something that already happens in the business: a webhook from a form or CRM, an inbound call, a WhatsApp message. Nobody has to open a dashboard for the agent to begin.
- Orchestrate
- n8n or FastAPI owns the run. It routes the request, calls slow tools asynchronously so a caller is not left waiting in silence, and retries a failed step instead of dropping the request.
- AI agent
- The model is one replaceable part of the system. Claude, GPT, Gemini or a local Ollama model does the reasoning, picked for the job through model routing rather than wired in for good.
- Actions
- Results land where the team already works: a CRM record, an API call into an internal system, a reply on WhatsApp.
What the agent draws on
- Knowledge
- Answers come from the client’s own documents through pgvector retrieval with reranking, and each answer cites the source it came from.
- Tools
- Function calling and MCP give the agent a fixed list of things it may do. Each one is an explicit call that shows up in the trace.
- Memory
- Redis keeps the session, so a conversation, a call or a returning user picks up where it left off instead of starting again.
- Evaluate and ship
- RAGAS evaluations and tracing check the answers before anything ships; Docker images then go out through CI/CD to AWS EC2.
Career
Career, in dated steps
Education first, then each role with its dates and what the work involved, ending with MiniAI Labs. Employer names are my own career history, as listed on my LinkedIn profile.
– Education
B.E. Computer Science and Engineering
Annapoorana Engineering College, Salem
– Role
Cloud & Linux Administrator (DevOps focus)
Nux Software Solutions (partner client of Airavad), Coimbatore
- Managed Linux infrastructure (RHEL, Ubuntu, Rocky) and containerised workloads on Kubernetes
- Modular Terraform for EC2, S3, RDS and VPCs with security policies applied by default
- Lambda + EventBridge automation and Python incident-diagnostics scripts on LLM APIs
- Prometheus and Grafana monitoring and alerting to improve uptime
– RolePromoted
AI DevOps & MLOps Engineer
Airavad Web Solution, Salem
- MLOps pipelines on AWS SageMaker and MLflow for versioned training, testing and deployment
- Prometheus, Grafana and Alertmanager paging for Kubernetes and SageMaker monitoring
- CI/CD for AWS EKS with CircleCI, GitHub Actions and Terraform, moving to rolling releases designed to avoid downtime
- AI-assisted monitoring triage with Ollama and AWS Bedrock (FastAPI, LangChain) and n8n pipelines across Git, Slack, CloudWatch and Lambda
– Role
AI Automation & DevOps Engineer
Hodos 360, Bengaluru (multi-client agency)
- Production voice AI agents (Retell AI, Vapi, ElevenLabs) for lead qualification and support, calling backend APIs in real time
- Async, non-blocking tool calls and session resumption, designed to avoid dead air and mid-call drops
- RAG pipelines on pgvector with reranking on OpenAI and Claude, answering from client knowledge with cited sources
- Multi-agent automations in n8n and Python/FastAPI connecting CRM, WhatsApp and internal tools
- Multi-tenant stack on AWS EC2 (Docker, Nginx, PM2, CI/CD) with rolling releases designed to avoid downtime, tracing and prompt versioning
- Next.js client sites with technical SEO and AEO: server-rendered HTML for Googlebot, JSON-LD (BreadcrumbList, FAQPage), i18n routing fixes, a pre-deploy SEO audit script
– presentFounder
Founder
MiniAI Labs
Skills
Nine domains, one working stack
The tools I use day to day, grouped the way a build uses them: models and retrieval at the centre, the automation and voice layers around them, and the cloud, data and search work that keeps a system running and findable. Choose a domain to see its tools.
Claude (Anthropic)
Tools and methods
- Claude API
- Tool use
- Structured outputs
- Prompt caching
- MCP
- RAG with source citations
AI automation
Tools and methods
- n8n
- LangChain
- LangGraph
- LlamaIndex
- Multi-agent orchestration
- Function calling
- Webhooks
- REST APIs
- Async pipelines
LLM and RAG
Tools and methods
- OpenAI GPT
- Claude
- Gemini
- Ollama
- Embeddings and reranking
- Hybrid retrieval
- pgvector
- Pinecone
- Qdrant
- ChromaDB
- Prompt engineering
- Guardrails
ML and MLOps
Tools and methods
- AWS SageMaker
- MLflow
- Training and testing pipelines
- Versioned deployment
- Model monitoring
- AWS Bedrock
- Vertex AI
- Google Colab
Voice AI
Tools and methods
- Retell AI
- Vapi
- ElevenLabs
- Whisper STT / TTS
- Gemini Live and Realtime APIs
- Async tool calling
- Session resumption
LLMOps and evals
Tools and methods
- RAGAS
- Langfuse
- LangSmith
- Tracing
- Prompt versioning
- Hallucination handling
- Retry logic
- Model routing
Cloud and DevOps
Tools and methods
- AWS (EC2, Lambda, S3, RDS, EKS, IAM, CloudWatch)
- GCP (Cloud Run, Vertex AI)
- Docker
- Kubernetes
- Terraform
- GitHub Actions
- CircleCI
- Nginx
- PM2
- Cloudflare
Backend and data
Tools and methods
- Python
- FastAPI
- Node.js
- TypeScript
- Next.js
- PostgreSQL
- Redis
- Supabase
- DynamoDB
- Prometheus
- Grafana
- Alertmanager
- Linux
SEO and AEO
Tools and methods
- Technical SEO
- Crawlability and indexation
- SSR for Googlebot
- Schema.org JSON-LD
- i18n URL routing
- AEO / GEO answer content
- Semrush Site Audit
Builds
Five systems, built end to end
Products I designed, built and deployed myself. Each card shows what the system does and the path a request takes through it. They are part of my own track record, not MiniAI Labs client projects, and nobody they were built for is named.
- Built by Vijay
Autonomous SEO blog engine
Picks a title from live trend data every hour; Gemini writes the article and its featured image; each draft passes an SEO filter and score check before it is posted.
Flow
- Trends
- Title
- Gemini article
- Gemini image
- SEO score check
- Publish
StackGemini API (text + image) · Hourly scheduler · SEO scoring
- Built by Vijay
Voice-controlled website agent
Gemini Live gives native speech-to-speech; tool calls open the page or section the visitor asks for; a thinker/talker split and async tools are designed to avoid dead air, and session resumption outlasts the ~10-minute limit.
Flow
- Voice command
- Gemini Live
- RAG memory
- Navigate site
- Show and speak
StackGemini Live API · FastAPI · Async tools · Redis · RAG
- Built by Vijay
Multi-app AI command chatbot
One agent publishes blogs, starts video jobs, books appointments, checks servers and builds reports — Whisper voice notes, 24-hour memory and language detection; one EC2 deployment serves several clinics.
Flow
- Chat or voice note
- Agent router
- Blog · Video · Appointments · Server health · Reports
StackTelegram · GPT-4o · Whisper · Redis · PostgreSQL
- Built by Vijay
AI video generation SaaS
One interface over several video models: fallback routing hides vendor differences in audio control and duration caps, and a job queue archives every asset to S3 before the providers’ expiry.
Flow
- Prompt
- Model registry
- Veo / Kling / Sora with fallback
- ElevenLabs voice
- S3 archive
StackNext.js · Python pipeline · Job queue · S3
- Built by Vijay
Scraping and SEO automation
Playwright crawlers and Node.js audit scripts check rendering, metadata and JSON-LD before each deploy.
Flow
- Crawl
- Extract
- Audit rendering, meta, schema
- Score
- Report
StackPlaywright · Node.js · Python · GitHub Actions
Film
The founder film
This page as a short silent film: the agent wiring, the career steps, the skill domains and the five builds. The captions are part of the picture, so it reads the same with the sound off, and the full transcript sits beside it.
Your device asks for reduced motion, so the film shows its poster until you choose to play it.
Transcript
Silent film with burned-in captions. Select a line to jump to it.
Credential
Credential and education
Certification
AWS Certified Solutions Architect – Associate
Amazon Web Services
A personal certification held by Vijay. It is not a certification of MiniAI Labs as a business.
Education
- –Diploma in Computer Science and EngineeringThiagarajar Polytechnic College
- –B.E. Computer Science and EngineeringAnnapoorana Engineering College, Salem
Working together
How engagements with Vijay work
One person from the first call to the handover, with the commercial terms written down before any paid work starts.
- The free 30-minute scoping call is with me, and so is every call after it.
- I write the proposal myself: scope, fixed price, timeline and an estimate of the running costs on your accounts.
- I build with demos you can try along the way, and I tell you when the first working version is ready, which starts your 14-day money-back window.
- At handover you get the documentation and admin control of every account in your name; after that, quoted changes or a monthly plan if you want one.
FAQ
Questions about Vijay
What did Vijay work on before he founded MiniAI Labs?
Infrastructure first. From May 2024 he ran Linux and Kubernetes workloads and wrote Terraform for AWS at Nux Software Solutions, a partner client of Airavad. In May 2025 Airavad promoted him to AI DevOps & MLOps Engineer, and from November 2025 he built voice agents, retrieval pipelines and multi-agent automations at Hodos 360, a multi-client agency in Bengaluru.
Which certification does Vijay hold, and does it cover MiniAI Labs?
He holds AWS Certified Solutions Architect – Associate, issued by Amazon Web Services. It is a personal certification in designing systems on AWS. It is not a certification of MiniAI Labs as a business, and the studio does not present it as one.
What did Vijay study?
Computer science, twice: a Diploma in Computer Science and Engineering at Thiagarajar Polytechnic College from 2016 to 2019, then a B.E. in Computer Science and Engineering at Annapoorana Engineering College, Salem, from 2021 to 2024.
Has Vijay put voice agents and retrieval pipelines into production use?
Yes, at Hodos 360. He built voice agents on Retell AI, Vapi and ElevenLabs that qualify leads and handle support while calling backend APIs during the call, and retrieval pipelines on pgvector with reranking that answer from a client’s own knowledge and cite their sources.
Are the five builds on this page MiniAI Labs client projects?
No. They are part of Vijay’s own track record: systems he designed, built and deployed end to end, described by what they do and how a request moves through them. No one they were built for is named, and none of them is presented as MiniAI Labs client work.
What does Vijay check before an AI system is released?
That its answers hold up and that it can be shipped safely. Retrieval answers are scored with RAGAS evaluations, runs are traced, prompts are versioned, and releases go out through CI/CD as Docker images so a new version replaces the old one without taking the system down.
Is the agent wiring on this page what every project gets?
It is the shape most of his builds take, not a fixed product. A proposal maps the same pieces onto your systems: what starts a run, which orchestration and model fit, what the agent may read and do, and how it is evaluated before release. A simple workflow may need only some of them.
Can we check Vijay’s background before the first call?
Yes. His LinkedIn profile lists the same roles and dates shown here. The strategy call is with him too, so you can ask about any role, tool or build directly before you commit to anything. Personal phone numbers are not published; the booking link and the contact form both reach him.
Next step
Talk to the engineer who will build it.
Book a free 30-minute strategy call with Vijay. Bring one workflow; you will leave with a view on scope, the systems involved and a sensible first step.
Free 30-minute call · no commitment