Jaswanth Samba

Certified in Cognitive Project Management in AI (PMI-CPMAI)

AI Engineer building production LLM agents and the data platforms behind them.

Turning enterprise data into working systems since January 2020, across manufacturing, retail, and healthcare. Based in Ontario, Canada.

View projects Download resume Email me

employees the HR AI agent was designed to serve: ~50,000; employees the HR AI agent was designed to serve
people trained and led across 10 countries: 26; people trained and led across 10 countries
automated executive dashboards (CEO, CFO, CXO): 15+; automated executive dashboards (CEO, CFO, CXO)
spend tracked across 3,000+ Tesco stores: £1.5B; spend tracked across 3,000+ Tesco stores

Professional AI work · Musashi Auto Parts

Production LLM systems inside a global manufacturer

Described at the level of architecture and impact. Diagrams are generic on purpose: no internal names or data.

01
HR AI agent
Production-scale HR assistant designed to serve ~50,000 employees.
Hybrid RAG with multi-layer memory.
- FastAPI
- Voyage AI embeddings
- Pinecone
- Supabase
- Claude
1. 1Employee question
2. 2FastAPI service
3. 3Hybrid retrievalVoyage + Pinecone
4. 4Memory layersSupabase
5. 5Claude answer
02
Musashi One GPT: enterprise policy assistant
Employees sign in with Microsoft SSO (Entra ID) and ask policy questions.
A router sends each question to specialist agents for HR, Finance, Quality, Maintenance, and Compliance.
- LangChain
- Pinecone
- Supabase
- Claude
- Entra ID
1. 1SSO sign-inEntra ID
2. 2Router
3. 3Specialist agentsHR · Finance · Quality · Maintenance · Compliance
4. 4Policy answer
03
Security-trimmed RAG ingestion pipeline
Azure Functions pipeline that ingests documents into Azure AI Search.
Custom index schema and hybrid search; results respect each user's document permissions.
- Azure Functions
- Azure AI Search
- Hybrid search
1. 1Documents
2. 2Azure Functionschunk + embed
3. 3AI Search indexcustom schema
4. 4Permission filter
5. 5User results
04
Knowledge-handoff Teams bot
An AI assistant in Microsoft Teams that answers questions about my systems and processes, so knowledge doesn't leave with the person.
- Voyage embeddings
- Pinecone
- Claude Sonnet
- Microsoft Teams
1. 1Teams message
2. 2RetrievalVoyage + Pinecone
3. 3Claude Sonnet
4. 4Answer in Teams

Data engineering and analytics

The data platforms the agents stand on

SAP to Microsoft Fabric lakehouse: Pipelines moving SAP ERP data into a medallion lakehouse with Change Data Capture and incremental loads. Translated ABAP data-model logic into SQL stored procedures and views.
Cloud migration: Moved on-premises SSIS and SQL Server ETL into Microsoft Fabric.
Real-time IoT telemetry: Python and REST ingestion of global device telemetry into Fabric pipelines feeding live manufacturing dashboards.
Executive Power BI: 15+ fully automated CEO, CFO, and CXO dashboards across 10 countries, including advanced DAX, forecasting measures, and a custom Deneb/Vega org chart visual.
Global enablement: Subject-matter expert for Fabric company-wide; led AI and data engineering training in Japan and trained a team of 26 across 10 countries.

Personal projects

Four worlds, built on my own time

Scroll through them, then open one for the full case study.

01 / 04Private voice assistant

Luna

A voice assistant that runs entirely on my laptop, with no cloud calls.

llama3.2:3b on an RTX 3050 Ti laptop GPU: 59 tok/s; llama3.2:3b on an RTX 3050 Ti laptop GPU
Airflow runs succeeded across 3 pipelines: 7 / 7; Airflow runs succeeded across 3 pipelines
of logged utterances embedded for memory: 100%; of logged utterances embedded for memory

To doDemo video and dashboard screenshotTo doMedian "hear you to first sound" latency

Read the case study

1Wake word"Luna"
2Speaker recognition
3Llama 3.2 3Blocal model
4Kokoro speech

Everything on this path runs locally.

02 / 04Swipe-to-answer trivia game

Doom Trivia

An endless trivia game whose questions come from an AI pipeline built on 40 topics and 2,000 subtopics.

curated questions shipped: 637; curated questions shipped
game modes: 6; game modes
passing tests (2026-09-13): 51; passing tests (2026-09-13)

To doQuestion validation write-up

Read the case study

Doom Trivia Tough mode: a question card asking which civilization built Machu Picchu, with four answers in the corners of the screen.

Doom Trivia home screen: a glowing lantern mascot on a mossy forest rock under a large mushroom, with the heading "Follow your curiosity" and New Game and Load Game buttons.

03 / 04AI job application agent

The Job Hunter

A multi-user agent that finds jobs, ranks them against your resume, and writes tailored cover letters.

To doReal counts from SupabaseTo doRow level security re-enabledTo doDashboard screenshot with dummy data

Read the case study

1Indeed listingsPlaywright
2Fit scoreClaude vs. resume
3Cover letter PDF
4TrackerSupabase + React

Every application is tracked end to end.

04 / 04Voice-note appointment booking

Voice Dental Scheduler

A patient leaves a voice note and the appointment books itself.

automated tests across 2 files (results pending): 8; automated tests across 2 files (results pending)

To doRun the tests and record resultsTo doDemo with a fake patient

Read the case study

1Voice note
2Whispertranscription
3GPTextracts details
4Google Calendar
5Twilio SMSconfirmation

From voice note to confirmed booking without a receptionist.

Experiments

Smaller things, honestly framed

TQQQ LSTM direction predictor
A live pipeline pulling hourly TQQQ data (5,000+ bars), engineering 15 technical indicators, training an LSTM in TensorFlow, and storing everything in SQLite.
Honest finding: Neither the hourly nor the daily model beat a naive "always up" baseline (ROC-AUC below 0.5). Short-term direction of a leveraged ETF wasn't predictable with this setup, which made it a useful lesson in evaluating models against baselines.
Silver futures moving-average strategy
Backtested a crossover strategy on the full history of silver futures while working through Python for Algorithmic Trading.
DentalAI Hub
An early React and Firebase dashboard for dental clinics with AI chat and call handling.
AI-native BI desktop app (in progress)
LangGraph agent, DuckDB, React/Tauri.
To confirmConfirm whether to include this in-progress project.

Experience

Where I've done the work

Aug 2022 – present
Business Data Engineer (AI Engineer)
Musashi Auto Parts · Waterloo, ON
- Build production LLM agents and RAG systems (see professional AI work).
- Lead Fabric data engineering and data science enablement for a team of 26 across 10 countries; led training in Japan.
- Migrated on-premises SSIS/SQL Server ETL to Microsoft Fabric; built SAP-to-lakehouse pipelines with CDC and incremental loads.
- Built real-time IoT telemetry ingestion and 15+ automated executive dashboards.
Jun 2021 – Jul 2022
Business Intelligence Analyst, Property Insights
Tesco · India
- Built a Tableau dashboard tracking £1.5B in capex and opex across 3,000+ UK stores and 60 KPIs for the UK Head of Property, coordinating 7+ teams.
- Built an asset condition model forecasting capital needs for a £17M three-year budget.
- Rebuilt a 15-report library from Excel into Tableau.
Jan 2020 – May 2020
Business Intelligence Analyst
Compass Group (Crothall) · Canada
- Built a Power BI unitizing analysis for 15 hospitals that saved the equivalent of 36 full-time roles in scheduling.
- Call centre analysis that set staffing shifts and cut costs by 20%.
- Cut report execution time by 50% across dashboards used weekly by 40+ hospitals.

Education, certification and skills

Foundations and toolkit

Education

University of Waterloo
Master of Management Sciences (MMSc), Data Analytics and Project Management
2018 – 2019
BMS Institute of Technology (VTU)
Bachelor of Engineering, Electronics and Communication
2013 – 2017

Certification

PMI

Cognitive Project Management in Artificial Intelligence

May 2025

AI and LLMs: Claude
OpenAI
RAG (hybrid, security-trimmed)
Multi-agent systems
LangChain
LangGraph
Embeddings (Voyage)
Pinecone
Azure AI Search
Local LLMs (Llama)
Speech (Whisper, Kokoro)
TensorFlow
Data engineering: Microsoft Fabric
Azure Data Factory
Spark / Spark SQL
Databricks
SAP HANA
SSIS
Apache Airflow
Medallion architecture
CDC
Real-time streaming
Backend and apps: Python
FastAPI
SQL
Supabase
Firebase
Next.js
React
Flutter
Playwright
Docker
REST and GraphQL
Analytics: Power BI
DAX
Deneb/Vega
Tableau
KQL
Power Query

Contact

Building something with AI and data? Let's talk.

Email: jashsamba@gmail.com
Phone: +1 (647) 769-5588

GitHub
LinkedIn
Resume (PDF)
To doUpwork link (optional)

AI Engineer building production LLM agents and the data platforms behind them.

HR AI agent

Musashi One GPT: enterprise policy assistant

Security-trimmed RAG ingestion pipeline

Knowledge-handoff Teams bot

Luna

Doom Trivia

The Job Hunter

Voice Dental Scheduler

TQQQ LSTM direction predictor

Silver futures moving-average strategy

DentalAI Hub

AI-native BI desktop app (in progress)

Business Data Engineer (AI Engineer)

Business Intelligence Analyst, Property Insights

Business Intelligence Analyst

Education

Certification

Building something with AI and data? Let's talk.