Open to new opportunities

Heema Maniar.

Analytics Engineer  ·  Strategic Data Analyst  ·  KPI Strategy & AI-Augmented BI

10+ years partnering with Finance, Marketing, Sales, Customer Success, and Operations to build the KPI frameworks, self-service dashboards, and data pipelines that give every team one source of truth — and the confidence to act on it.

San Francisco Bay Area, CA maniarheema@gmail.com 10+ Years Experience

About

Data is my craft.
Decisions are my output.

Analytics engineer and strategic data analyst with 10+ years turning complex data into business strategy across SaaS, fintech, and tech. I work across data engineering, analytics, and cross-functional leadership — building the pipelines, KPI frameworks, and self-service tools that let every team answer their own questions from one trusted source of truth.

I've partnered directly with Finance, Marketing, Sales, Customer Success, Operations, Product, and HR to define metric ownership, standardize KPI calculations, and deliver go-to dashboards that leadership actually uses to make decisions. At a SaaS company, I automated reporting pipelines that cut manual effort by about 30% and used LLMs to surface support ticket patterns at scale. At a global fintech platform, I built an automated dashboard system that lifted team productivity by roughly 35% and designed predictive fraud detection models.

I use AI and prompt engineering to speed up every part of the analytics workflow, from data exploration to narrative generation. Currently building independently across BI, dbt, and agentic analytics.

BI & Visualization

Looker Tableau Google Data Studio Periscope Salesforce

Data Engineering

dbt Fivetran Airflow BigQuery Snowflake

Programming

SQL Python LookML R GitHub

Cloud & AI

GCP Google ADK Gemini Claude API LLMs / NLP Prompt Engineering

Strategy & Collaboration

KPI Governance Self-Service BI Stakeholder Management Single Source of Truth

Featured Work

Dashboards & Data Products

Hands-on analytics builds, from operational dashboards to end-to-end dbt projects. More work from past roles is being rebuilt as shareable portfolio pieces.

Operations Analytics · dbt + ML/LLM Build

Operations Productivity Intelligence

Synthetic Salesforce-style case data for a financial-services ops team, modeled end to end (CSV → BigQuery → dbt → Data Studio). Mines ticket patterns with ML and Claude, stands up a knowledge base, and benchmarks a classic ML classifier against an LLM for triage. Every number is measured from the data, not asserted.

↑ 24% lower handle time on KB-addressable tickets (26.0 → 19.7 min)
BigQuerydbt Data StudioPythonClaudescikit-learn

Data & Analytics · Analytics Engineering

Revenue Intelligence (Olist)

Real Brazilian e-commerce data (99K orders, 9 relational CSVs) modeled in BigQuery + dbt — staging → intermediate → marts, tested and documented — into a live, code-built Evidence dashboard. Ships with a KPI dictionary (single source of truth), a leadership decision brief, and Claude-tagged themes/sentiment over 1,500 Portuguese reviews — surfacing decisions like “lateness is a carrier-transit problem, not seller dispatch.”

↑ Live dashboard · 99K real orders · 1,500 AI-tagged reviews
BigQuerydbt EvidencePythonClaude API

AI-Powered Apps

AI-Powered Analytics Apps

Multi-agent AI systems — live, deployed, and production-grade.

Multi-Agent AI · Hackathon · Fivetran Track

🌉 GoldenGate Retail AI

AI co-pilot for Bay Area shopping mall general managers. Routes natural-language questions to three specialist agents (a Data Unifier on BigQuery + Fivetran MCP, a Tenant Diagnoser for lease risk and rent-to-sales ratios, and an Action Recommender running ARIMA_PLUS 30-day forecasts) over ~1.5M synthetic transactions across 13 malls. Every answer is traceable to the SQL behind it.

↑ Google Cloud Rapid Agent Hackathon · June 2026
Google ADKGemini 3 Flash BigQuery MLFivetran MCP StreamlitCloud Run

Self-Improving AI · Arize Phoenix Evals

🔬 EvolvBI

Self-improving SQL analytics agent. Ask questions in plain English; every interaction is traced in Arize Phoenix and scored by three evals, including a grounding check that catches fabricated numbers. A one-click loop reads the failure patterns and rewrites the agent's own prompt, shown as a live red/green diff and saved to BigQuery so the gains persist across runs.

↑ Live on Cloud Run · self-improving eval loop · June 2026
Google ADKGemini 3 Flash Arize PhoenixBigQuery ML StreamlitCloud Run

Learning

Certifications & Education

Continuously upskilling at the frontier of data and AI.

Database & Data Analytics

UCSC Silicon Valley Extension · 2018

Earned

Bachelor of Computer Engineering

University of Mumbai · 2012

Earned

dbt Fundamentals

dbt Learn

Completed

Databricks Fundamentals

Databricks Academy

Completed

Introduction to Generative AI

Google Cloud Skills Boost

Completed

Let's work together

Get in Touch.

Open to Senior / Lead Analytics Engineer, Data Analyst, and Strategic Data Analyst roles where data drives decisions across the business.

Send an Email View LinkedIn → GitHub
linkedin.com/in/heema-maniar github.com/heemaniar maniarheema@gmail.com