SHUBHANKAR_TIWARI
SYSTEM STATUS: ONLINEColumbia MS AI '27 · Finance & OpsPatent Pending (2026)

Shubhankar
Tiwari

Software Engineer with 4+ years building enterprise microservices at Bank of America • M.S. in AI at Columbia University (Expected Dec 2027).

Building LLM evaluation tools, entropy-based generation controls, and drift-aware ML pipelines.

0+
CashPro Corporate Users
0M+
Searches Served (SME)
0+ Yrs
Enterprise Backend SWE
Top 4.1%
Kaggle Expert (#2,441)
Live Telemetry
Latency 18ms•Baseline AUC 0.975•Policy gates active
Explore AI Systems Experience & Patent Resume (PDF) CLI Terminal
Background

About & Core Focus

Shubhankar Tiwari

I've always been drawn to understanding how complex systems actually work. Whether it's enterprise software or machine learning, I care most about how things behave under pressure, why they break, and how to make them more reliable.

I spent four years at Bank of America building and maintaining backend services for CashPro, their enterprise digital banking platform. That taught me a lot about the trade-offs in designing software that has to keep evolving without falling over. I picked up multiple Global Recognition Awards along the way and got involved in AI initiatives within the org.

I bring that same systems thinking to AI. Over the past year, I've been building LLM evaluation tools, reliability frameworks, and model governance pipelines, while getting deeper into Python, PyTorch, and Transformers.

Now I'm at Columbia doing an M.S. in Artificial Intelligence (Finance & Operations concentration) to go deeper on the research side.

0+ Years
Enterprise Production SWE
#2,441
Kaggle Expert / 59.7K
94.4%
B.Tech CSE GPA (SRM IST)
34
Public ML Notebooks
Engineering Philosophy

Systems Mindset Applied to Artificial Intelligence

[01]

Reliability Over Hype

Evaluating model behavior token by token. Catching failure modes during generation, not after user delivery.

[02]

Deterministic Governance

Building policy controllers, CI gates, and business-loss promotion metrics that prevent silent model degradation.

[03]

Production Rigor

Four years of enterprise engineering at Bank of America across 17+ production releases and 13 CashPro services serving 550K+ users.

[04]

Continuous Adaptation

Integrating real-time PSI & KL drift detection with shadow deployment to evaluate recovery before promotion.

Career Track

Professional Experience & Patent

4+ years building and scaling enterprise systems at Bank of America and fintech. Went from production engineering to SME to AI patent inventor.

4+ Years Enterprise Production

Served as Subject Matter Expert (SME) for CashPro Search, the primary search platform across Bank of America's digital banking ecosystem. It serves 550,000+ corporate users across 40,000 companies with 18 million+ searches since launch.

Owned OpenShift deployments, CI/CD configuration, release validation, and production stabilization across 17+ release cycles. First point of contact for cross-team production escalations and architectural decisions.

Contributed to modernization initiatives including JDK 17, Spring 6, containerization, and cloud enablement across 13 CashPro services, improving team standing in vulnerability, security, and modernization metrics via Checkmarx and SonarQube scans. Orchestrated AI adoption initiatives for 50+ CashPro colleagues.

Named inventor on a pending Bank of America patent application (2026) in AI for financial services. Led 10+ team discussions to refine the idea, architecture, and presentation.

JavaSpring BootREST APIsOpenShiftCI/CDSQLCheckmarxSonarQube

Bank of America · Global Recognition Awards

Gold Award

Global Recognition • BofA

Backend delivery, release ownership, & cross-team SME lead

Silver Award

Global Recognition • BofA

Cloud enablement & OpenShift service containerization

Bronze Award

Global Recognition • BofA

Production incident stabilization & rapid defect triage

Systems & Models

Featured Engineering Systems

Software and AI systems I've built for evaluation, monitoring, and governance in production environments.

BROWSE ALL 10 PROJECTS
Flagship AI ArchitectureSep 2025 - Dec 2025 • Completed

AI Reliability Platform for LLM Evaluation

Built a full-stack system to monitor and control LLM generation at the token level, with adaptive instability detection and real-time intervention.

Entropy-aware routing, adaptive regeneration policies, and behavioral workflows across 213 prompts spanning 5 behavioral categories
100% harmful-output catch rate on 43 adversarial prompts across 9 holdout rounds on fine-tuned mT5; extended testing to Qwen 2.5-7B, Llama 3 8B, and Mistral v3
9-category guardrail classifier with escalation logic. 0% leakage and 0% hallucination across 213 evals
PythonMLOpsLLMControl SystemsSystem Design
Read Case Study View Code
100%
Adversarial Catch
82%
Instability Drop
0.0%
False Positives
4
Design Iterations
Interactive Architecture SimulationLIVE

LLM Generation Entropy & Governance Pipeline

Streaming Ingestion

Active Focus

Extracts live probability distribution P(x_t) and calculates Shannon entropy H(x) per generated token in real-time.

H(x) = -sum(P(w) * log2(P(w)))
43 Tokens
Active Stream
AI Reliability Platform for LLM Evaluation
Sep 2025 - Dec 2025Completed

AI Reliability Platform for LLM Evaluation

Built a full-stack system to monitor and control LLM generation at the token level, with adaptive instability detection and real-time intervention.

•Entropy-aware routing, adaptive regeneration policies, and behavioral workflows across 213 prompts spanning 5 behavioral categories
•100% harmful-output catch rate on 43 adversarial prompts across 9 holdout rounds on fine-tuned mT5; extended testing to Qwen 2.5-7B, Llama 3 8B, and Mistral v3
PythonMLOpsLLM
Drift-Aware Fraud Detection: ML Lifecycle & Model Governance
Feb 2026 - Jul 2026Completed

Drift-Aware Fraud Detection: ML Lifecycle & Model Governance

Built an end-to-end ML system that detects data drift, evaluates model degradation, and governs retraining decisions.

•Complete ML system loop: Frozen Dataset → Failure-mode metrics (AUC degradation, PSI drift) → Δ Analysis → Policy Gate (Business Loss < 0) → CI Promotion
•Maintains 0.975 AUC-ROC baseline with business-loss-aware promotion gates using 10:1 FN/FP cost weighting
FastAPIscikit-learnNext.js
Conservative Auto-Regeneration Policy for AI-Generated Financial Narratives
2025Completed

Conservative Auto-Regeneration Policy for AI-Generated Financial Narratives

Production validation system for AI-generated outputs in financial services: threshold optimisation, conservative AND escalation policy, priority-scored human review queue, and audit-grade provenance.

•Conservative AND policy: auto-regen only when both embedding similarity AND ROUGE-L fall below threshold. Single-metric failure routes to human review
•Threshold calibration via bootstrap percentile sweep across labeled validation set
PythonNLPFinTech
Indian Desi Multilingual LLM: Training Pipeline
2025Completed

Indian Desi Multilingual LLM: Training Pipeline

End-to-end multilingual LLM training pipeline targeting Hindi/English code-switching. Dataset curation, LoRA fine-tuning, inference evaluation, and deployment packaging across 6 Kaggle notebooks.

•Canonical dataset curated from 3 complementary sources: chatbot dataset for tone, large-scale conversation corpus for diversity, sentence-pair dataset for structural grounding
•Unified schema normalising all sources into a clean, consistent format
PythonLLMNLP
Song Recommender System
2021Completed

Song Recommender System

ML-based workout song recommender using BPM and VADER sentiment analysis. Co-authored research with K-Means clustering on Billboard Top 100 to match songs to exercise intensity.

•Co-authored research project with Tanish Maheshwari (Presidency University)
•Lyrics extracted from Billboard Top 100 via Genius API and lyrics-extractor library
PythonMLNLP
Kaggle Portfolio
2020-PresentActive

Kaggle Portfolio

Notebooks Expert rank #2,441 / 59,663 (personal best #707). 34 notebooks, 11 datasets, 3 models, 1 competition entry. 10 bronze medals across ML, DL, NLP, Computer Vision, and regression.

•Notebooks Expert: current rank #2,441 / 59,663 | personal best #707
•34 published notebooks with 10 bronze medals
Data ScienceMLDeep Learning
Headlines Text Generation using LSTM
2021Completed

Headlines Text Generation using LSTM

Predicted and generated new headlines using NLP and LSTM networks.

•LSTM-based sequence model for natural language generation
•Trained on news headline corpus for next-word prediction
PythonNLPLSTM
Trigger Word Detection
2021Completed

Trigger Word Detection

Speech recognition system that detects the trigger word 'activate' using NLP and plays a chime sound upon detection.

•Detects trigger word 'activate' in audio stream using NLP
•Real-time speech recognition with chime activation on detection
PythonNLPSpeech Recognition
Face Recognition System
2020Completed

Face Recognition System

Built a face recognition system using OpenCV in Python to detect faces, eyes, and smiles, identifying individuals with confidence scores.

•Face detection with component recognition (eyes, smile)
•Person identification with confidence scoring against trained dataset
PythonOpenCVComputer Vision
Estadísticos: Stock Prediction & Classification
2020Completed

Estadísticos: Stock Prediction & Classification

Predicted Tesla stock prices using ML algorithms with descriptive, prescriptive, and predictive analysis. Also applied classification algorithms to PIMA India Diabetes dataset.

•Tesla stock price prediction using ML algorithms
•Descriptive, Prescriptive, and Predictive Analysis approaches
PythonMLClassification
Capabilities & Stack

Technical Focus & Technologies

AI & ML Systems

Core skills in this area

Evaluation pipelinesControl policiesDrift detectionFailure-mode metricsShadow deployment
Systems Readiness

Right now I'm focused on the overlap between ML research and production engineering: building evaluation tools, deployment governance, and reliability infrastructure.

Explicit Technology Stack

AI & Machine Learning
PythonPyTorchTransformersScikit-learnXGBoostKerasOpenCVTensorFlow
ML Systems & Reliability
LLM EvaluationAI ReliabilityModel GovernanceBehavioral TestingLLM GuardrailsPrompt EngineeringPSI & KL Drift
Enterprise Backend
JavaSpring BootREST APIsSQLOpenShiftJenkinsAnsibleLinuxJUnit

Credentials & Recognition

Kaggle Notebooks Expert

Rank 2,441 / 59,663 · Personal Best #707 · 10 Bronze Medals

M.S. AI · Columbia University

Expected Dec 2027 · Finance & Operations Concentration

ML & Deep Learning Certified

Kaggle & Deep Learning Specialization · 2025

IELTS Certification

Band 8 / 9 · Dec 2025

Chronology

Engineering & Academic Timeline

From undergraduate computer science foundations to enterprise banking systems, Kaggle global rank, and AI graduate research at Columbia University.

2018 → 2027 Evolution

MS in Artificial Intelligence

Aug 2026 – Dec 2027

Columbia University · New York, NY · Concentration: Finance & Operations

Graduate student in Artificial Intelligence, focusing on ML systems and AI reliability. Building on four years of enterprise software engineering and hands-on work with LLM evaluation frameworks.

Software Engineer I A · SME

2024 - 2026

Bank of America

Promoted to Subject Matter Expert for CashPro Search, serving 550,000+ corporate users. Owned deployments, CI/CD, and release validation across 17+ cycles. Named inventor on a pending Bank of America patent application (2026).

Software Engineer I B

2023 - 2024

Bank of America

Increasing ownership of backend services through feature development, production support, and service reliability. Diagnosed and resolved 20+ production incidents spanning application and infrastructure layers.

Apprentice

2022 - 2023

Bank of America

Built foundational experience in enterprise software engineering by contributing to the CashPro Search platform using Java, Spring Boot, and Agile delivery practices.

Kaggle Notebooks Expert

2022 - Present

Top 4.1% Globally

Notebooks Expert ranked #2,441 / 59,663 (personal best #707). 34 notebooks and 10 bronze medals across ML, DL, NLP, Computer Vision, and regression.

B.Tech CSE

2018 - 2022

SRM IST · 94.4% GPA

Graduated with 94.4% GPA in Computer Science. Built foundation in algorithms, data structures, and software engineering. Active in college tech/coding clubs and tech fests.

Interactive CLI

Systems Telemetry & Terminal

shubhankar@systems-cli:~
>welcome

SHUBHANKAR TIWARI // INTELLIGENT SYSTEMS TERMINAL v2.5.0

Type 'help' to see available commands or click quick suggestions below.

QUICK COMMANDS:
>
GET IN TOUCH

Let's Build Reliable Systems

Open to opportunities in ML Systems, AI Infrastructure, Reliability Engineering, and Platform Architecture.

Send Email
Resume (PDF)GitHubLinkedInKaggleColumbia
st3907@columbia.edu•tiwarishubhankar@gmail.com