SR
Menu

Cover sheet · New York, NY

Sariya Rizwan

Data Engineer & Business Intelligence Analyst | AI-enabled data systems

4+ years solving ambiguous business problems in technology, healthcare, energy, finance, product and operations. I pair quantitative modeling and SQL/Python analytics with cross-functional work, so recommendations get adopted and dashboards, forecasts and AI-enabled tools keep running.

View projects Resume (PDF) Get in touch

$120K

annualized savings at PSEG

6h → 5 min

fleet-planning processing time at PSEG

$200K+

client value delivered at ASH-TECH

30%

reduced onboarding time at WebScape

01 · About

Where data, analytics and AI meet

I start with open-ended business questions, define the KPIs, model the scenarios and turn the results into tools and recommendations leaders actually use.

Today I own analytics workstreams at Natera, including Voice of Patient, AI Disposition and Twilio/Calabrio reconciliation. I also lead a modernization workstream for a mission-critical gas-electric incident notification system at Consolidated Edison.

Earlier, at PSEG and ASH-TECH Solutions, I built forecasting tools and financial models that turned ambiguous operational and client questions into decisions.

Outside work I build LLM agents, ML infrastructure and evaluation tooling. My published research includes GRAFT, a gRPC-routed agent framework (ACM AI and Agentic Systems, 2026), and Branch-Commit-Validate for autonomous red-team agents (Springer SAM, 2025).

02 · Experience

Roles and timeline

Consolidated Edison (ConEd)

Oct 2026 – Present

New York, NY (Hybrid)

In progress

Junior Business Analyst – GIS

Owns a cross-functional modernization workstream for a mission-critical gas-electric incident notification system, from requirements through validation and delivery.

  • Facilitate workshops with GIS, control-center, support, architecture and engineering leaders to synthesize requirements and map current and future states.
  • Build consensus on 200-ft/24-hour proximity logic, centralized architecture, monitoring/security and implementation sequencing.
  • Translate analysis into executive-ready BRD/FRD, process/data-flow, reporting, UAT, migration and rollout recommendations.
  • GIS
  • Requirements
  • Workshops
  • BRD/FRD
  • Process/data-flow
  • UAT

Natera

Jun 2026 – Present

San Carlos, CA (Remote)

In progress

Business Intelligence Analyst

Owns strategic analytics workstreams and delivers recommendations used by Operations, Product, Finance, Sales, Support and leadership.

  • Own analytics across Voice of Patient, PC Performance, AI Disposition, Salesforce transfer reporting, communication hold/release and Twilio/Calabrio reconciliation.
  • Run deep dives, forecasts, variance, funnel, A/B and pre-post analyses and scenario modeling with SQL and Python.
  • Reconcile conflicting data, automate recurring decision tools, and influence roadmap priorities through executive narratives, QA/UAT and training.
  • SQL
  • Python
  • Snowflake
  • Databricks
  • BigQuery
  • Power BI
  • Tableau
  • Sigma
  • Excel
  • Salesforce
  • Twilio
  • Calabrio

Public Service Enterprise Group (PSEG)

Jun 2025 – May 2026

Edison, New Jersey

In progress

Technical Intern, Continuous Improvement

Solved an ambiguous fleet-planning problem with a custom forecasting tool and led continuous-improvement workstreams across operations.

  • Built Fleet Forecaster with React, Flask, PostgreSQL, SQL, Python and Excel/VBA to model fleet demand, capacity, cost and budget scenarios.
  • Reduced processing from 6 hours to 5 minutes and delivered approximately $120K in annualized savings.
  • Led continuous-improvement workstreams from discovery through implementation, automating workflows with Power Apps, Power Automate, Python and APIs.
  • React
  • Flask
  • PostgreSQL
  • SQL
  • Python
  • Excel
  • VBA
  • Power Apps
  • Power Automate
  • APIs

ASH-TECH Solutions

Jul 2023 – Aug 2024

Gujarat, India

Data Analyst – Business Intelligence & Financial Analytics

Led client-facing analytics and strategy engagements and managed BI and data workstreams from requirements through adoption.

  • Turned analysis into executive recommendations that delivered $200K+ in client value.
  • Built performance and financial models covering growth drivers, profitability, ROI, forecasting and scenarios.
  • Integrated Salesforce, HubSpot, NetSuite, APIs and cloud data, and automated analytics with SQL, Python, dbt and Airflow.
  • SQL
  • Python
  • dbt
  • Airflow
  • Salesforce
  • HubSpot
  • NetSuite
  • APIs

WebScape

Jul 2021 – Jun 2023

Mumbai, India

Business Intelligence Analyst – Data & Systems Analytics

Partnered with clients to diagnose business and operational issues and owned several delivery workstreams at once.

  • Standardized reusable processes that reduced onboarding time by 30%.
  • Presented recommendations on customer, product, conversion, retention, churn, sales, marketing and financial performance.
  • Used advanced SQL, Python and BI tools to investigate root causes and supported UAT and training.
  • SQL
  • Python
  • BI tools

03 · Projects

Blueprints in code

Each project is drawn as the thing it does: a pipeline, a gate, a graph, a log. Thirteen projects, from forecasting to supply-chain security.

LLM agents

AI Travel Agent

A tool-calling travel agent on the Anthropic API that searches flights and hotels, checks weather and builds day-by-day itineraries. It runs as a CLI chat or a FastAPI endpoint with multi-turn conversations.

Each turn is capped at 8 model calls and a 60-second deadline, and the agent returns an honest fallback instead of fabricated results.

  • Python
  • Anthropic API
  • FastAPI
  • Uvicorn
  • uv

Source code for AI Travel Agent

RL evaluation

Process Reward Model RL Environment

An RL environment in which an LLM agent trains a Process Reward Model end-to-end. The model is graded against an Outcome Reward Model baseline and a PRM800K-style reference on a held-out MATH slice.

The real PRM scores 0.770 in the demo, while the reward-hacking submissions score 0.000.

  • Python
  • NumPy
  • SymPy
  • pytest

Source code for Process Reward Model RL Environment

NLP tooling

JobPilot

A local-first job-hunting assistant that discovers jobs, parses descriptions with spaCy, scores resume fit transparently and drafts tailored resumes and outreach. It never auto-applies or auto-sends.

Uses a transparent 7-component estimated match score, and every outreach draft requires manual review.

  • Python
  • spaCy
  • Uvicorn
  • Alembic
  • Jinja2
  • APScheduler

Source code for JobPilot

Forecasting

hierforecast

Hierarchically reconciled energy demand forecasts using MinT shrinkage reconciliation, scored on pinball loss. Regional forecasts sum exactly to the national figure and the upper tail stays accurate.

Independent forecasts disagreed by up to 612 MW; after reconciliation the gap is float round-off.

  • Python
  • MinT reconciliation
  • Quantile forecasting
  • unittest
  • GitHub Actions

Source code for hierforecastLive demo for hierforecast

MLOps

modelcanary

Progressive delivery for ML models: a FastAPI gateway serves ONNX models in shadow and then canary. Sequential tests decide automatic promotion or rollback.

Moves from a shadow stage to canary stages (e.g. 10% then 50%) with sticky salted-hash routing.

  • Python
  • FastAPI
  • ONNX Runtime
  • scikit-learn
  • Prometheus
  • uv

Source code for modelcanary

ML pipelines

dagcache

A content-addressed, parallel pipeline runner for reproducible ML workflows. It hashes code, params and data, explains why a stage is stale and traces artifact lineage.

SHA-256 cache keys cover code, params, input content and env vars, and AST normalization ignores comments.

  • Python
  • YAML
  • S3
  • mypy
  • uv

Source code for dagcache

Agent infrastructure

replayagent

A single-node, Temporal-style durable execution runtime for LLM agents. Event-sourced SQLite history and deterministic replay let a workflow resume after a crash without re-running finished steps.

Every state change is one lease-fenced SQLite transaction, with zero runtime dependencies.

  • Python
  • SQLite
  • Ollama
  • GitHub Actions

Source code for replayagent

ML security

picklescope

A static scanner for ML model files that inspects pickle, PyTorch, Keras, safetensors, GGUF and ONNX without deserializing them. It reports findings as text, JSON or SARIF.

Checks every import against an allow-list policy, with format-specific checks for documented attacks.

  • Python
  • PyTorch
  • Keras
  • safetensors
  • GGUF
  • ONNX

Source code for picklescope

Supply-chain security

squatwatch

An offline detector for typosquats, dependency-confusion exposure and risky install hooks in npm and PyPI dependencies. It outputs SARIF for CI.

Compares names against a snapshot of about 12,000 top PyPI projects and 14,000 high-impact npm packages.

  • Python
  • PyPI
  • npm
  • SARIF
  • uv

Source code for squatwatch

Detection rules

sigmaforge

A Sigma detection-rule compiler that parses rules into a typed expression tree and renders Splunk SPL, Elastic EQL, Microsoft KQL and SQLite SQL, checked by property-based differential tests.

Each backend consumes the same expression tree, and the SQL output runs in SQLite to check the others.

  • Python
  • Sigma
  • Splunk SPL
  • Elastic EQL
  • KQL
  • SQLite

Source code for sigmaforge

Log parsing

evtxkit

A from-scratch Rust parser for Windows EVTX event logs with a chunk-parallel JSONL pipeline, CRC verification, damaged-file recovery and Sigma-style field mapping for detection work.

Matches the reference parser on all 4,805 records of 9 public sample logs.

  • Rust
  • JSONL
  • Sigma-style mapping
  • CRC32

Source code for evtxkit

Audit integrity

tamperlog

A hash-chained audit log that adds Merkle checkpoints with externally anchored roots.

Detects re-chained truncation that plain hash chaining silently accepts.

  • Python
  • SHA-256
  • Merkle checkpoints
  • Standard library only

Source code for tamperlog

Rule quality

yaraci

A CI harness for YARA rule quality that lints rules, proves they fire on fixtures, scans goodware for false positives, profiles scan cost and diffs rule versions.

Gates pull requests on behaviour, not just syntax, with JUnit and SARIF reports.

  • Python
  • YARA
  • JUnit
  • SARIF

Source code for yaraci

04 · Skills

Parts list

The tools I reach for, grouped the way a build list is grouped.

  1. Programming

    • Python
    • SQL
    • VBA

  2. Data platforms & engineering

    • Snowflake
    • Databricks
    • BigQuery
    • PostgreSQL
    • dbt
    • Airflow

  3. AI & ML tooling

    • Anthropic API
    • LLM tool calling
    • FastAPI
    • ONNX Runtime
    • scikit-learn
    • spaCy

  4. Analytics & modeling

    • Quantitative modeling
    • Forecasting
    • Scenario modeling
    • A/B and pre-post analysis
    • Data modeling
    • KPI development
    • AI-enabled solutions

  5. BI & visualization

    • Power BI
    • Tableau
    • Sigma
    • Excel
    • Dashboard development

  6. Frameworks & automation

    • React
    • Flask
    • Power Apps
    • Power Automate
    • APIs

  7. Business systems

    • Salesforce
    • HubSpot
    • NetSuite
    • Twilio
    • Calabrio

  8. Delivery

    • Requirements gathering
    • Business analysis
    • UAT
    • Process improvement

05 · Education & research

Study and papers

Education

In progress

Master of Business Administration (MBA)

Business Analytics (STEM) · Pace University

New York, NY · Ends May 2026

Published research

GRAFT, a gRPC-routed agent framework

ACM AI and Agentic Systems, 2026

Branch-Commit-Validate for autonomous red-team agents

Springer SAM, 2025

Paper at EDULEARN25

EDULEARN25

06 · Contact

Let’s build the next data system

Looking for someone who can turn messy data into forecasts, models and AI-enabled tools that teams rely on? Reach out by email or connect on LinkedIn.

sariyarizwan25@gmail.com