Open to $190K+ Lead DS Roles

Girijesh Singh

Lead Data Scientist  ·  Applied AI Architect

Building production AI where hallucinations carry legal consequences. 6 years · $10M+ impact · 2,000+ users.

2,000+
Users
30K+
Daily Queries
$10M+
Impact
6 yrs
Depth
SCROLL
PydanticAIFastMCPAzure OpenAIPyTorchTransformersXGBoostSHAPFastAPIPySparkAzure Cosmos DBOpenTelemetryArize PhoenixHierarchical RAGGPT-4oAzure Service BusScikit-learnPythonSQLDockerCitation ValidationMulti-Agent SystemsNLPFine-tuningPydanticAIFastMCPAzure OpenAIPyTorchTransformersXGBoostSHAPFastAPIPySparkAzure Cosmos DBOpenTelemetryArize PhoenixHierarchical RAGGPT-4oAzure Service BusScikit-learnPythonSQLDockerCitation ValidationMulti-Agent SystemsNLPFine-tuning
PydanticAIFastMCPAzure OpenAIPyTorchTransformersXGBoostSHAPFastAPIPySparkAzure Cosmos DBOpenTelemetryArize PhoenixHierarchical RAGGPT-4oAzure Service BusScikit-learnPythonSQLDockerCitation ValidationMulti-Agent SystemsNLPFine-tuningPydanticAIFastMCPAzure OpenAIPyTorchTransformersXGBoostSHAPFastAPIPySparkAzure Cosmos DBOpenTelemetryArize PhoenixHierarchical RAGGPT-4oAzure Service BusScikit-learnPythonSQLDockerCitation ValidationMulti-Agent SystemsNLPFine-tuning

What I Bring

What I Bring to the Table

Three things that separate a production AI architect from someone who can run a notebook.

01

Zero-to-One GenAI Architecture

I don't just prompt-engineer; I build deterministic, multi-agent microservices from scratch. When off-the-shelf tools like LangChain fail at scale, I design custom orchestration layers (PydanticAI, FastMCP) that actually work in production.

02

High-Performance System Optimization

I bridge the gap between Data Science and Data Engineering. I optimize vector indices (Azure Cosmos DB IVF) to slash query latencies from minutes to seconds, circumventing hundreds of thousands in cloud scale-out costs.

03

Measurable Enterprise Impact

My architectures don't just live in notebooks. I've directed teams of 10–14 engineers to deploy systems that drive $10M+ in operational efficiency and process tens of thousands of queries daily across 4 global regions.

Technical Philosophy

Why I Stopped Using LangChain for Enterprise RAG

Moving a Generative AI prototype to a regulated production environment exposes the severe limitations of standard wrapper libraries. Within 30 days of deploying standard RAG on highly interlinked, 100+ page insurance documents, I diagnosed catastrophic context collapse and citation failure.

My solution wasn't a larger context window - it was a better architecture. I engineered a custom runtime schema transpilation layer and a hierarchical node retrieval engine from scratch. The result? A system that captures complex cross-references without hallucinations, dynamically scaling to thousands of users while keeping cloud costs flat.

“The best GenAI architecture isn't the one with the most features - it's the one that can provably tell you exactly why it gave the answer it did, every single time.”

By the Numbers

Real Impact at Scale

Every number below is earned, not estimated from a demo.

0+
Active Users
Insurance adjusters · US · UK · AU · EU
0K+
Daily AI Queries
Multi-carrier · Multi-tenant
0×
Latency Reduction
Hours of work → under 30 seconds
$0M+
Annual Savings
In insurance adjuster time
0
Global Regions
US · UK · Australia · EU
0×
Throughput Gain
More queries · same hardware

Featured Work

What I've Built

Production systems. Published research. Shipped products.

🏗️
PRODUCTION · NDA

CoverAI: Zero-Hallucination Retrieval Engine

Lead Data Scientist · 2024–Present · Confidential Employer

Problem: Off-the-shelf LangChain RAG failed catastrophically on 700+ page insurance policies — hallucinating citations with legal consequences.

Build: Custom hierarchical JSON tree retrieval with runtime schema transpilation. Deterministic citation validation streams output character-by-character. New carriers onboard via config, zero code changes.

🔥 Scale: 30K+ daily queries · 2,000+ enterprise users
Speed: 10× latency (7.5s → 0.7s) · end-to-end under 30s
🛡️ Safety: 100% citation validation · zero hallucinated references
💰 Cost: $500K+ cloud scale-out costs avoided
PydanticAIFastMCPHierarchical Node RetrievalAzure Cosmos DB (IVF)MicroservicesReal-Time Citation ValidationFastAPIOpenTelemetryAzure OpenAI/GPT-4oAdobe PDF Services
10×
Latency
<30s
E2E Response
$500K+
Cloud Saved
2,000+
Active Users
30K+
Daily Queries
4
Regions
PRODUCTION

Autonomous Multi-Line Claims Routing

ML Engineer / Lead · 2018–2021

XGBoost pipeline for insurance claim triage. SHAP values for 100% regulatory audit trail. CTO: "No dedicated team has ever built anything like this." 45% cost reduction · $200K annual savings · 35% faster settlements.

XGBoostSHAPDjangoRabbitMQExplainable ML
🔤INTERNAL

PriML: Natural Language → SQL

Team Lead of 10 · 2018–2021 · Pre-LLM era

Fine-tuned Rat-SQL transformer. NL query → SQL → Plotly dashboard, self-serve. 87% accuracy on complex multi-table JOINs before LLMs existed.

Transformers (Rat-SQL)Fine-tuning87% AccuracyPlotlyPostgreSQL
PRODUCTION · NDA

FNOL Classification Agent System

Lead Data Scientist · 2024–Present

3-service microarchitecture (FastAPI + FastMCP + Azure Service Bus). PydanticAI agent generates type-safe Pydantic models from per-tenant schemas at runtime. 95% alignment · 2% hallucination rate.

PydanticAIFastMCPAzure Service BusDynamic Schema GenerationMulti-Tenant
📄PUBLISHED

Selective EEG Anonymization via Multi-Objective Autoencoders

PST 2023 · Copenhagen, Denmark · Lakehead University

Privacy-preserving ML for Brain-Computer Interfaces. Selective anonymization preserves clinical signal while eliminating re-identification vectors.

View on IEEE Xplore ↗
Multi-Objective AutoencodersEEG/BCIsPyTorchPST 2023

Side Projects

🎬OPEN SOURCE

DirectorAI

Browser-native AI video editor. Natural language → FFmpeg.wasm. TensorFlow.js face detection client-side. No uploads, no server, no privacy tradeoff.

React 19FFmpeg.wasmTensorFlow.jsVite
GitHub
🔌LIVE

AI Conversation Exporter

Export ChatGPT, Claude, Gemini conversations as TXT, Markdown, JSON, or HTML. Zero permissions. All processing is local. Chrome + Firefox.

Chrome/Firefox ExtensionManifest V3Zero Permissions
GitHub

Technical Arsenal

What I Work With

🤖LLM Orchestration
FastMCPPydanticAIHierarchical Node RetrievalCitation ValidationAzure OpenAI/GPT-4oCohere RerankingEmbedding Models
📊Core ML / AI
PyTorchTransformersXGBoostSHAPVision OCRNLP/Fine-tuningProphetScikit-learn
💻Languages
PythonSQL
☁️Data & Cloud
PySparkAzure Cosmos DB (IVF)Azure Service BusAWS (Lambda·SageMaker·S3)RabbitMQPostgreSQLMongoDBDocker
🚀Delivery
FastAPIOpenTelemetryArize PhoenixDjangoFlaskAdobe PDF ServicesTeam Leadership (10–14)

Experience

Career Timeline

Jan 2024 – Present

Lead Data Scientist

Primus Software Corporation

Waterloo, ON

Led cross-functional team of 10–14. Scaled enterprise AI to 2,000+ users globally. Resolved two production crises. Built zero-code carrier onboarding.

Promoted from Senior → Lead in 12 months

Jan 2023 – Jan 2024

Senior Data Scientist

Primus Software Corporation

Waterloo, ON

Diagnosed LangChain's fundamental limits on multi-document insurance policies. Designed hierarchical RAG architecture solo in 3 months. Resolved latency crisis: 2.5 min → 40 sec.

Sep 2021 – Apr 2023

M.Sc. Computer Science

Lakehead University

Thunder Bay, ON

Project-based Masters supervised by Dr. Garima Bajwa. Published privacy-preserving ML research at PST 2023, Copenhagen. Continued AI development at Primus concurrently.

Published at PST 2023 · Copenhagen

May – Aug 2022

Data Science Intern

Ciena

Ottawa, ON

PySpark pipelines, divisive clustering, manufacturing batch anomaly detection.

Jun 2018 – Dec 2022

ML Engineer → Senior ML Engineer

Primus Software Development

Noida, India → Canada (2021)

Built FNOL classification for Crawford & Company solo. Led PriML NL-to-SQL project (team of 10). 6 years of insurance domain expertise starts here.

CTO recognition + $1,000 bonus

Research

Publications

📚 Peer-Reviewed · IEEE · International Conference

Selective EEG Signal Anonymization using Multi-Objective Autoencoders

PST 2023 · Copenhagen, Denmark

Co-authored research on advanced autoencoder architectures for securing biological telemetry. Selective anonymization preserves clinical signal while eliminating re-identification vectors. Supervised by Dr. Garima Bajwa, Lakehead University.

View on IEEE Xplore ↗
📚 Peer-Reviewed · Springer

In-Memory Computation for Real-time Face Recognition

ICICT 2019 · Springer

Optimized edge-compute inference for computer vision on resource-constrained hardware. In-memory computation strategies significantly reduce latency for real-time face recognition workloads.

View on Springer ↗

Beyond the Code

Leadership & Communication

Data scientists who can communicate build better systems. The evidence is below.

🎤

Toastmasters International

Competitive public speaking training that directly informs how I present technical findings to non-technical stakeholders — executives, clients, and insurance carriers.

7× Best Impromptu Speaker
4× Best Evaluator
3× Best Prepared Speech
👥

Cross-Functional Team Lead

Ran day-to-day technical and delivery decisions for a team of 10–14. Direct stakeholder requirement gathering, refinement, and brainstorming. Second-most senior person on the team.

10–14 person cross-functional team
Multi-region, multi-carrier delivery
Client-facing requirement ownership

Let's Work Together

Open to the RightLead & Staff DS Roles

Available for $190K+ Base Compensation
Waterloo area or remote · Big Tech, AI-native, Enterprise AI
If your engineering bar is high and you need an architect who thinks in systems rather than just notebooks, let's talk.

singhgirijesh1996@gmail.comgithub.com/girijesh18LinkedIn

Currently based in Waterloo, ON · Open to remote-first teams