Abhishek Tiwari · ML Engineer

experiment report // 2022 - 2026

Abhishek
Tiwari

ML Engineer·Data Scientist ·AI Engineer

I build end-to-end machine-learning systems that ship - pipelines, serving, drift monitoring and the metrics to prove they work.

Selected work ↓ Résumé (PDF) ↗
~4yrs enterprise AI
850github commits
34public repos
4live ML apps
4xsuper squad award
model: abhishek.v4 - training run training
1.0 0.3 0.1 0 60 epoch loss 1.0 0.5
epoch 60/60 loss 0.150 acc 0.948
val_loss ↓val_acc ↑

01 - Overview

I turn data into systems that make decisions.

I'm a machine-learning engineer and data scientist. I build end-to-end ML systems - data pipelines, model serving, drift monitoring and CI/CD - and four of them run live today, each with experiment tracking and a test suite behind it.

By day I'm a Lead Assistant Manager at EXL Service, where I've spent nearly four years on the enterprise-AI side: Agentic-AI proofs of concept and GenAI demos for internal and external stakeholders. Promoted twice; recognised four times for project impact.

I'm now looking for ML Engineer, Applied AI or Data Scientist roles at product companies and AI-native startups - places where shipping measured, reliable models is the job.

02 - Stack

Instruments & methods.

core ml / python
Pythonscikit-learnXGBoost / LightGBMAdaBoostSHAPpandas / NumPyimbalanced-learn
mlops & deployment
MLflowDagsHubDockerGitHub ActionsFastAPIStreamlitpytestdrift monitoring
data & sql
SQL (MariaDB / MySQL)SQLiteMongoDB Atlasdata wranglingPlotly / SeabornA/B testing
techniques
SMOTE / ImbPipelineSMOTETomekIsolation ForestGridSearchCVcross-validation
ai & enterprise
Appian BPMAgentic AIGenerative AILLM conceptsRPA
tooling & auth
GitJupyterPyJWT / bcryptReportLabPydantic
currently learning
LangChainRAGPyTorch

$ dashed = in progress · everything else is shipped in competition work.

03 - Experience

Depth over breadth - 4 years, 1 company, 2x Promotions.

2026 -
now

Lead Assistant Manager - R&D & AI POC Lab

EXL Service · Noida (WFH)
  • Currently on the demo & experimentation track - building Python ML proofs-of-concept and data-science experiments to validate ideas before they go into production pipelines.
  • Designing and presenting end-to-end ML demos for stakeholders - from raw data through model training, evaluation and a Streamlit front-end.
  • Running rapid experiments across models and datasets: tracking every run in MLflow, comparing metrics, and packaging the best results into shareable demo apps.
  • Python
  • ML engineering
  • Streamlit
  • Insurance
2024 -
2025

Assistant Manager - R&D & AI POC Lab

EXL Service · Noida (WFH)
  • Led an Agentic-AI POC for short-term-disability claims on Appian +15% productivity - as an early-access collaborator feeding direct feedback to Appian's product team.
  • Architected a Digital Quality Assurance enterprise MVP (calendar management + audit forms) that cut delivery time 10% and removed manual errors across the team.
  • Built 15+ GenAI / Agentic-AI demos for internal and external stakeholders; earned 4x Super Squad Award across two years.
  • Agentic AI
  • GenAI
  • R&D
  • POC / MVP
2022 -
2023

Senior Executive - Junior Developer

EXL Service · Noida (WFH)
  • Built the Appian layer for an anomaly-detection POC - consumed ML predictions via API into the database and surfaced results to business users through process models and interfaces.
  • Shipped a real-estate domain BPM app in a 9-person team: process models, SAIL interfaces, records and MariaDB integrations from the ground up.
  • Standardised reusable Appian component libraries, cutting recurring effort and onboarding time for new modules.
  • Appian BPM
  • SAIL
  • MariaDB
2021

Data Analyst Intern

Suven Consultants & Technology · Mumbai (WFH)
  • Performed EDA on meteorological datasets with pandas and NumPy, surfacing seasonal patterns and anomalous trends.
  • Built an MNIST digit-recognition model with scikit-learn across the full pipeline; stakeholder-ready visuals in Matplotlib and Seaborn.
  • Python
  • Pandas
  • EDA
  • scikit-learn

04 - Selected work

Six builds, read like figures.

Each is an end-to-end system with results attached. Repos are public; four run live; demos are on YouTube.

FIG.01

PhishGuard

live

End-to-end MLOps pipeline for phishing URL / domain detection.

features
30 url/page
pipeline
6 stages
selection
multi-model + GridSearchCV
promotion
F1-gated ≥ 2%
drift
KS-test
serving
REST + Streamlit

click a stage to inspect it - its design diagram loads straight from the repo · high-level design ↗

PythonFastAPIscikit-learnMLflowDagsHubDockerGitHub Actions
↗ Live app ⎇ Repo ▶ Demo
Six sequential stages pass typed artifacts down the pipeline: ingestion, validation, transformation, multi-model training with GridSearchCV, KS-based data-drift detection and an F1-gated promotion step that only ships a new model when it beats production by a set margin. Predictions are served over a REST API with a Streamlit dashboard on top.
FIG.02

System Performance Analyzer

live

Real-time host telemetry with anomaly detection and auth.

monitors
cpu/ram/net/proc
anomaly
Isolation Forest + z-score
health
0-100 (A-F)
auth
JWT + bcrypt
reports
on-demand PDF
charts
interactive Plotly
PythonStreamlitPlotlyscikit-learnSQLitePyJWTbcryptReportLab
↗ Live app ⎇ Repo ▶ Demo
Tracks CPU, RAM, swap, battery, network and per-process metrics in real time. Isolation Forest plus a z-score check flag anomalies; a daily 0-100 health score rolls up to an A-F grade. Role-based admin access is gated with JWT and bcrypt, and any view exports to a one-click PDF report.
FIG.03

Telco Customer Churn

live

Churn prediction with a cost model bolted on.

auc-roc
0.8403
recall
0.7727 (+41.6% SMOTE)
net roi
₹19.4L / cycle
cv auc
0.8480 (5-fold)
explain
SHAP top-20
customers
7,032
PythonAdaBoostSMOTESHAPStreamlitSQLitepytestGitHub Actions
↗ Live app ⎇ Repo ▶ Demo
An AdaBoost classifier flags churn before customers leave, protecting ₹19.4L of net ROI per cycle over 7,032 customers. SMOTE lifted recall by 41.6% and SHAP narrows inference to the top-20 features. A drift-monitoring dashboard, a SQLite audit trail and a pytest-backed CI pipeline keep it production-honest.
FIG.04

Thunderstorm Forecasting

live

Next-day thunderstorm classifier on atmospheric soundings.

task
next-day binary
model
QDA + SMOTETomek
features
sounding indices
tracking
MLflow
PythonQDASMOTETomekscikit-learnMLflowStreamlitimbalanced-learn
↗ Live app ⎇ Repo ▶ Demo
QDA quietly outperformed tree-based models on this dataset. SMOTETomek handles severe class imbalance by combining oversampling with Tomek-link cleaning, and every experiment is tracked in MLflow so the model-selection story stays reproducible.
FIG.05

Retail Data Insight

CLI analytics over 180K+ e-commerce transactions.

records
180K+
charts
8 publication-grade
reports
3-level PDF
scope
revenue · geo · customer
PythonpandasMatplotlibSeabornmodular OOPPDF generation
⎇ Repo
A command-line analytics tool that runs EDA across 180K+ transactions - revenue trends, top customers, best-selling products and geographic patterns - then emits 8 publication-quality charts and auto-generates Level 1-3 PDF reports, from summary stats up to geo-intelligence.
FIG.06

Financial Tracker

Menu-driven CLI finance tracker, built to be production-readable.

storage
CSV-backed
tracks
income / expense
reports
net balance
errors
custom exceptions
PythonCSVOOPloggingCLI
⎇ Repo
A clean, modular CLI tracker: add income/expense transactions, browse history and generate net-balance reports. CSV persistence, rotating logs and custom exception classes keep it readable and maintainable despite being terminal-only.

05 - Recognition

Ranked, awarded, certified.

Honours & competitions

  • Global rank #8 - HR AnalyticsAnalytics Vidhya · 2023
  • GATE 2025 qualified - AIR 155MN · IIT · 2025
  • 4x Super Squad AwardEXL Service · 2024-25
  • Global rank #67 - Power Plant AIDockership · 2021
  • AWS Machine Learning ScholarshipUdacity · 2021
  • Bertelsmann Technology ScholarshipUdacity · 2020
  • 2x Kaggle bronze notebook medalKaggle · 2021
  • 1st prize - Vigyaan'19Aavartan, NIT Raipur · 2019
  • Finalist - NEI Idea FactoryIdea Factory, NBC · 2021

Certifications

  • Appian Certified Associate Developer
    Appian Corporation · valid through May 2027
  • Machine Learning - Stanford University
    Coursera / Andrew Ng · 2020
  • EXL Hyperautomation Training
    Great Learning · 2022
  • 3x Udacity Nanodegrees
    Data Analytics · AI with Python · Predictive Analytics

06 - Writing

Field notes from production ML.

Long-form write-ups of things that actually happen to models after they ship - drift, leakage, the gap between offline metrics and reality.

$ latest from Medium · all posts ↗

07 - Contact

Let's build something measurable.

Or reach me directly

click to copy