trace · dalbirsingh-matharu
trace_id 8879703613-dm region thane, in status available

Every LLM demo works.
I ship the 10% that keeps it working at 2am.

Production AI engineer who's shipped multi-agent architectures and RAG pipelines that hold up under real traffic — not just notebooks. Built the AI review system running inside Longani's Word/PowerPoint add-ins today, and co-architected the SQL Assistant at Worter Technologies before that. Registered IP holder in AI-powered SQL generation, and the kind of engineer who reads the LangChain source closely enough to file a bug against it.

identity.json
{
  "name": "Dalbirsingh Matharu",
  "role": "AI Engineer",
  "specializes_in": ["agentic orchestration", "RAG", "LLMOps"],
  "currently": "Longani Consulting, LLC",
  "ip_holder": "SW-19859/2024",
  "education": "B.E. Data Science, CGPA "9.3
}
career_runtime calculating…
runtime / production
Experience
AI Engineer
Longani Consulting, LLC
● running Dec 2025 → present
  • Led development and delivery of a production AI review system for Office.js add-ins on Azure, coding 7+ features and 100+ automated checks, including asynchronous backend APIs and a custom rules engine.
  • Reduced AI-powered check latency to under 12 seconds in a live production codebase by designing LLM orchestration pipelines with feature-driven routing, PostgreSQL-backed workflow state, and iterative prompt testing and evaluation to improve check accuracy.
<12s p95 latency 100+ automated checks 7+ shipped features
Generative AI Engineer
Worter Technologies Pvt Ltd
✓ completed Jun 2025 → Nov 2025
  • Co-architected and deployed a multi-agent LLM-powered SQL Assistant using FastAPI on Google Cloud Platform, designing a custom orchestration pipeline with SQL validation, automatic retries, and LLM-based evaluation, achieving ~90% SQL accuracy.
  • Resolved schema retrieval issues across 400+ tables by building a RAG-based pipeline with ChromaDB, cutting response time to under 8-10 seconds via pruning, SQL caching, and access-control guardrails.
~90% SQL accuracy 400+ tables indexed 6–8s response time
side quests / shipped for fun
Projects
multi-agent

Data Analyst Agent

Upload a dataset, ask a question in plain English. A LangGraph pipeline cleans it, validates, retries on failure, runs EDA and root-cause analysis, generates charts, and profiles the data — end to end.

Hours of manual analysis → minutes.

LangGraphLangSmithGeminiStreamlit
↗ view source
multimodal

L.U.M.I.S — Multimodal AI Assistant

A real-time assistant that watches video, audio, and mouse activity to understand what you're doing and offer context-aware help, without you having to explain your screen to it.

Gemini 1.5 ProGPT-4LangChainOpenCV
↗ view source
mlops

Forecasting Platform — MLOps & Vertex AI

Multivariate sales forecasting with feature engineering and root-cause analysis, MLflow-tracked training across Linear Regression, Random Forest, and XGBoost.

R² > 0.99, MSE < 3, sub-50ms p95 serving.

Vertex AIDockerMLflowGitHub Actions
↗ view source
agentic ml

Machine Learning Agentic System

Give it a dataset and a question in plain English — it cleans and validates the data, automatically branches into supervised or unsupervised ML, trains and evaluates the model, then explains the result in plain language.

Auto-detects classification vs. regression vs. clustering, then builds, trains, and runs the model itself.

LangChainLangGraphGemini Proscikit-learn
↗ view source
upstream / off the clock
Open source & leadership
BUG

Filed and reproduced a bug in LangChain core

Reported a KeyError in PIIMiddleware affecting hash and mask redaction strategies — with a minimal, self-contained repro a maintainer could run as-is, in langchain-ai/langchain, the framework his own production systems are built on.

↗ issue #35647
LEAD

Software Lead, Modified Auto Club — Autonomous Vehicle System

Led a 5-person team to build an autonomous vehicle with real-time object detection, avoidance, and turn navigation using Raspberry Pi, Arduino, OpenCV, and sensor fusion, hitting ~85% real-world accuracy.

2nd place at Exault, a college project competition.

registered
Intellectual property
copyright · s.q.l.a.g.e
SQL Automated Generation and Execution
A registered, copyrighted approach to AI-powered SQL generation — not just a line on a résumé.
Copyright ID · SW-19859/2024
credentials / receipts
Certifications & GitHub

Certifications

Machine Learning, Data Science & Generative AI with PythonUdemy
Agentic AI Bootcamp — LangGraph & LangChainUdemy
Getting Started with Deep LearningNVIDIA
Power BI for Business IntelligenceUdemy

GitHub achievements

🦈
Pull Shark ×2
Quickdraw
🎯
YOLO
stack
What's under the hood
Languages
PythonJavaScriptSQL
Generative AI
Large Language Models (LLMs)Prompt Engineering Retrieval-Augmented Generation (RAG)AI Agents LLMOpsModel Context Protocol (MCP) Agent EvaluationGuardrails
AI Frameworks & ML Tools
LangChainLangGraphLangSmith MLflowPyTorchScikit-learn
Backend & APIs
FastAPIREST APIsAsync Processing API IntegrationPydantic
Cloud & DevOps
Google Cloud PlatformVertex AI AzureDockerGit/version controlGitLab CI/CD
Database & Vector DBs
MySQLPostgreSQLNeo4jFAISSChromaDB
Data Engineering & Analytics
PySparkApache SparkAzure DatabricksAzure Data LakesPower BI
AI Platforms & Models
OpenAIClaudeGoogle GeminiHugging Face
foundation
Education

B.E. in Data Science

A.P. Shah Institute of Technology, Thane — 2021 to 2025.

CGPA 9.3

High School

Rao Junior College of Science — 2019 to 2021.

95%