head 0.0 · previous-token
Agentic Solutions
Multi-agent systems that plan, use tools, and hold up in production, not just in the demo.
- LangChain
- LlamaIndex
- CrewAI
- MCP
- +3
Hemz · my local time
> generate(prompt="who builds production AI?")
Enterprise AI Engineer
shipping agents that survive production.
Token inspector
sequence · 7 tokens
logprob -.44 · perplexity 1.065
HoverTap a token in the headline (or step with the arrows) to see what else the model considered.
Tagline sampler
softmax(logits / T)- › shipping agents that survive production..50
- building the observability that makes agents trustworthy..24
- turning promising demos into dependable products..16
- watching every thought an agent has..06
- making machines think, one prompt at a time..02
- turning coffee into tokens since 2018..01
- arguing with LLMs so production doesn't have to..01
balanced: mostly sensible, occasionally fun.
01About
Demos are easy. Production is the point.
For 8+ years I've shipped agentic AI systems, the observability platforms that make them trustworthy, and the MLOps & LLMOps frameworks that keep them running at enterprise scale across retail, healthcare, and CPG.
My favorite territory is the gap between a promising demo and a dependable product. That's where the hard problems live: evaluation, observability, guardrails, and the unglamorous engineering that turns a clever model into something a business can trust.
Under it all is plain curiosity about the technology, and about how it reshapes the industries and people around it. It's what keeps me experimenting, writing, and shipping.
Profile card
Hemachandran Dhinakaran
Enterprise AI Engineer
- Current role
- Enterprise AI Engineer, Tredence Inc
- Experience
- 8+ years, since 2018
- Focus
- Agentic Solutions · Observability Platforms · MLOps & LLMOps
- Industries
- Retail · Healthcare · CPG
- Previously
- Machine Learning Engineer, Atos
- Machine Learning Practitioner, Tata Consultancy Services
- Education
- Certificate in Machine Learning, Stanford Online
- Bachelor of Engineering in Electronics & Communication, Sri Sairam Institute of Technology
02What I build
Three heads, one model.
The three disciplines that decide whether an AI system survives contact with real users, each drawn as the attention pattern it most resembles.
head 1.3 · induction
Observability Platforms
Tracing, evaluation, and guardrails that make every step an agent takes visible and debuggable.
- OpenTelemetry
- Arize Phoenix
- Langfuse
- SigNoz
- +2
head 2.7 · attention-sink
MLOps & LLMOps
Paved roads from experiment to production: CI/CD, retraining, deployment, and drift monitoring.
- Databricks
- Azure ML
- Azure DevOps
- CI/CD
- +3
03Experience
The career, as a trace.
8+ years, one long-running span. Open a span to read its attributes.
career.run()3 spans · 8y 3m · Jun 2018 → now status: in-flight
tcs.machine_learning_practitioner3y 6mMachine Learning Practitioner at Tata Consultancy Services, Jun 2018 to Dec 2021 (3y 6m)
Machine Learning PractitionerBuilt a self-heal automation API in Python: a Random Forest model that reads ticket parameters straight from the ticketing tool and triggers end-to-end remediation, cutting manual effort by 40%.
- span.start
- Jun 2018
- span.end
- Dec 2021
- span.kind
- full-time
- attributes
- PythonMachine LearningAutomationRandom Forest
atos.machine_learning_engineer1y 8mMachine Learning Engineer at Atos, Dec 2021 to Aug 2023 (1y 8m)
Machine Learning EngineerDesigned MLOps platforms on Databricks with end-to-end monitoring. Built data-centric AI pipelines with CI/CD handling end-user claims for a health-tech giant, and reverse-engineered legacy statistical SQL models into ML pipelines that cut business losses from claims.
- span.start
- Dec 2021
- span.end
- Aug 2023
- span.kind
- full-time
- attributes
- MLOpsDatabricksAzure MLCI/CDPython
tredence.enterprise_ai_engineer3y 1mEnterprise AI Engineer at Tredence Inc, Aug 2023 to present (3y 1m)
Enterprise AI EngineerBuilding enterprise AI products end to end: multi-agent systems, LLM-powered observability platforms, and production-grade MLOps & LLMOps frameworks serving CPG, retail, and healthcare clients.
- span.start
- Aug 2023
- span.end
- in-flight
- span.kind
- full-time
- attributes
- Agentic AILLMOpsMLOpsPythonAzureDatabricks
pre-training
- Stanford Online · Certificate · Machine Learning2020 — 2021
- Sri Sairam Institute of Technology · Bachelor of Engineering · Electronics & Communication2014 — 2018
04Selected work
Projects, embedded.
Every project plotted in a latent space, clustered by domain. Newer work glows brighter. Pick a point to open it.
Agents & Tools
notionwiki · 2026
NotionWiki
A Notion wiki your AI assistant can understand
An open-source CLI that turns a Notion workspace into a flat, indexed markdown wiki an AI assistant can actually reason over: immutable raw sources, agent-curated synthesis on top, OS-native scheduled sync, and a force-directed graph view.
- AI Assistants
- Notion API
- Python
- CLI
Observability
observent · 2026
Observent
Claude Code plugin for AI observability
A Claude Code plugin and Agent Skills that wire observability into multi-agent AI apps in one command: it detects your agent framework, instruments traces to Phoenix, Langfuse, or SigNoz, and validates the span hierarchy end to end.
- AI Observability
- Claude Code
- OpenTelemetry
- Agents
Agents & Tools
harness-mcp · 2025
Harness MCP
Reusable private AI toolkit
A private, reusable AI toolkit built on the Model Context Protocol: write a capability once and plug it into any MCP-aware client, without your data ever leaving your walls.
- MCP
- AI Tools
- Python