Blog · Tech
Understanding Vector Databases
12 min read
neuralwithmohit
Kuala Lumpur, Malaysia
Mohit Kumar Dubey
Head of AI, ML & Engineering · Entermind Malaysia
I build AI-first products, and the engineering organizations that ship them.
35+
engineers led across AI/ML, backend, mobile & web in a 500+ employee company
3×
delivery output at flat headcount via company-wide LLM automation
45%
lift in AI brand citation scores (GEO platform, GXS · Grab Group)
25%
loan-disbursal conversion uplift (BuddySmart agentic RAG)
30–40%
cut in LLM API cost via model routing & caching (DailyTatva RAG v2)
23%
reduction in lending default rates (credit-risk models)
Leadership
What I do
I lead AI and engineering at the level where a roadmap has to survive both a board conversation and a code review. Over 16+ years I've built and scaled engineering organizations from a handful of people to 35+ across AI/ML, backend, mobile and web, inside a 500+ employee company. I own engineering P&L (licenses, hiring, appraisals) and set the operating model that lets teams ship enterprise AI predictably: technical review gates, architecture sign-off, and delivery prioritization across concurrent engagements.
I carry that from strategy to production. At VerSe Innovation, India's largest short-video and news platform at 350M+ users, I owned the AI/ML roadmap across content and lending and turned it into business outcomes: a 25% loan-conversion uplift, 23% lower default rates, and 30–40% lower inference cost through model routing and cost governance. Today I run AI, ML & Engineering at Entermind, delivering enterprise and government AI across SEA and MENA, where I drove 3× delivery output at flat headcount and act as the primary technical partner to executives from a bank's Chief Credit Officer to the Grab Group.
I still design the systems myself: agentic RAG platforms, multimodal voice digital twins, and an open-source guardrail LLM. I care about the difference between a demo and a system that holds at scale, which comes down to unit economics, AI governance, evaluation before deployment, and the failure modes that only show up in production.
How I operate
Core competencies
Leadership
I design engineering organizations and the technology portfolio behind them. That means scaling teams across AI/ML, backend, mobile and web, developing the people in them, and leading cross-functionally from strategy through delivery.
Business
I own engineering P&L and treat AI product strategy as a commercial lever: supporting enterprise sales, enabling go-to-market, and building the strategic partnerships that turn capability into revenue.
Governance
I set the AI operating model (responsible-AI practice, governance gates, and enterprise adoption paths) and manage executive stakeholders from Chief Credit Officers to group leadership.
Expertise
Technical depth
AI & ML Engineering
- LLMs, RAG & Agentic RAG
- Multi-Agent Systems
- Fine-tuning (LoRA, PEFT)
- Evaluation & Guardrail Harnesses
- Prompt Engineering & Structured Outputs
- Recommendation & Credit-Risk Models
Agent & Serving Stack
- LangChain · LangGraph · CrewAI
- vLLM · SGLang · FastAPI
- LiteLLM gateway
- PyTorch · TensorFlow
- Voice: ElevenLabs, Tavus, LiveKit, Pipecat
Cloud & Platforms
- AWS Bedrock · SageMaker
- Vertex AI · Azure AI Foundry
- OpenAI · Gemini · Claude · Cohere
- Docker · Kubernetes · ECS · EC2 · CI/CD
- pgvector · Qdrant · Chroma · FAISS · Redis
Experience
Career
Entermind Malaysia
9 mos
Jan 2026 – Present
Kuala Lumpur, Malaysia
Head of AI, ML & Engineering
Jan 2026 – Present · 9 mosLead an 11-engineer team across backend, AI/ML, frontend and QA, and own the company-wide AI & engineering roadmap for enterprise and government clients across SEA and MENA.
- Drove a 3× increase in delivery output at flat headcount by identifying and automating recurring workflows with LLM pipelines.
- Improved pre-sales effectiveness 30% and cut claims and HR validation effort 45% by deploying enterprise RAG and conversational AI.
- Established the enterprise AI operating model: technical review gates, architecture sign-off, and delivery prioritization across concurrent engagements.
- Built a reusable GenAI capability layer (RAG ingestion, evaluation harness, guardrail scoring) so engagements start from proven assets, not scratch.
Selected work
Systems I've built
Entermind
Mindara, AI Employee for Enterprise Engagement
An always-on conversational AI colleague that replaces the intake form with adaptive follow-up questioning, surfacing real requirements and handing structured briefs to sales.
15% increase in solutioning lead calls
- LiteLLM
- Claude Opus
- Gemini
- Langfuse
- ElevenLabs
- LangGraph
Entermind
Rukun Ready, Guardrail LLM for Policy Validation
Fine-tuned Qwen2.5-32B to score LLM outputs against national policy principles and emit structured JSON compliance verdicts with safe rewrites. The firm's first open-source AI asset.
3K+ HuggingFace downloads
Products I've led
Shipped at scale
Dailyhunt
News & content · 350M+ users
Led AI/ML and platform architecture for India's largest local-language news app — personalized retrieval, reliability, and inference-cost governance.
30–40% lower LLM cost · 30% lower crash rate · 65% faster fixes
View on Play StoreJosh
Short video · 350M+ users
Led engineering across mobile, backend and ML for the short-video platform — streaming SDKs, recommendations, and reliability at massive scale.
99.8% crash-free · 90%+ launch-to-play · 50% faster playback
Knowledge base
What I write about
The proof of expertise is the content. Four bodies of work sit behind this page. Free and premium material live in the same lists, never partitioned off.
Blog→
Focused deep-dives on one concept, like vector databases, agent memory, or LLM gateway architecture, plus interview scenarios worked end to end.
AI System Design: Guidelines→
The methodology: how to approach designing any AI system. Requirements, evaluation harnesses, latency budgeting, failure modes, and when not to use an LLM.
AI System Design: Case Studies→
Complete architectures for real problems like RAG chatbots, email agents, and multi-agent platforms, each walked through in the same fixed order.
Cloud Component Comparison→
Equivalent services across AWS, Azure, and GCP side by side: the hard limits, pricing shapes, gotchas, and a verdict for each component.
Featured
Reading to start with
GuidelinesPremium
Designing an Evaluation Harness Before You Build
18 min read
Cloud Comparison
Lambda vs Azure Functions vs Cloud Functions
9 min read
Contact
Let's talk
Open to VP / Head of AI & Engineering roles.
Also open to consulting, collaboration, or a talk. The fastest way to reach me is on LinkedIn.
