Vikas Kumar | AI/ML Engineer

By the Numbers

Real Impact. Real Production.

These are not portfolio demos. These are outcomes from production AI systems I built under real accountability.

$3.2M

Logistics cost savings
in production

97%

Ontology accuracy
Knowledge Graph

96%

Document extraction
accuracy

23%

Delivery efficiency
improvement

3K+

Active users
Lawroom AI

Internships &
engagements

I treat prompt engineering as a systems design problem.

Context window architecture, output grounding, and hallucination mitigation are engineering constraints I treat with the same rigor as any systems design problem. I have solved all of these under real production pressure with real accountability.

I build at the hard end of the stack: RAG pipelines under retrieval pressure, Neo4j knowledge graph architectures over unstructured enterprise data, transformer-based document intelligence, and agentic systems that actually ship.

🧠 RAG & LLM Systems

Production-grade retrieval pipelines with vector search, context window management, and low-hallucination outputs deployed at scale using Azure OpenAI and LlamaIndex.

🕸️ Knowledge Graphs

Neo4j architectures modeling 500+ entities across semantic dimensions, integrated with LangChain to unify structured and unstructured enterprise data at 97% ontology accuracy.

⚡ Agentic AI

Multi-step autonomous systems with tool-use, persistent memory, and real-world API integration, designed to serve real users and not just run in notebooks.

📊 ML & Analytics

End-to-end pipelines from data engineering and feature design through model training, evaluation, and stakeholder-facing dashboards that drive actual decisions.

Academic Background

Education

Two fully funded degrees, awarded on merit across two continents.

Wake Forest University

MS in Business Analytics

2025 – 2026

100% Merit Scholarship · $85,000+

Current Work: Building RAG-powered career analytics system over Handshake data serving 500+ students and 40+ advisors using Azure OpenAI.

Courses: ML for Business, Python, Data Engineering, Optimization, Visualization, Career Management

Tools: Python, SQL, Power BI, Tableau, Scikit-learn, Azure OpenAI

Plaksha University

B.Tech in Artificial Intelligence

2021 – 2025

Full-ride Bharati Scholarship · $42,000+

Focus: Deep Learning, Econometrics, Time Series, NLP, Knowledge Representation, Design Thinking

Projects: LSTM predictive models, NLP chatbots, ESG-financial correlation, Knowledge Graphs

Tools: Python, PyTorch, TensorFlow, Selenium, Firebase, Git, Neo4j

Career Timeline

Work Experience

From production AI systems to research labs, a journey built on real outcomes.

Wake Forest University

AI Engineer

Mar 2026 – Present

North Carolina, USA

▸ Building a RAG-powered generative AI system that unifies Handshake, career portals, and institutional databases, enabling natural language querying over student employment and placement data using Azure OpenAI and vector embeddings, serving 500+ students and 40+ advisors.
▸ Designing ETL workflows to ingest, normalize, and index Handshake API data into a unified knowledge layer powering real-time LLM-driven analytics dashboards for faculty.
▸ Architecting prompt evaluation frameworks to reduce hallucination risk and ensure reliable structured data retrieval across diverse query types.

RAG Azure OpenAI LangChain PostgreSQL

Darden

Graduate Consultant

Oct 2025 – Present

▸ Built a Streamlit dashboard analyzing DoorDash and Uber Eats campaigns across 13,993 store-week observations and 109 stores, applying causal inference methods (DiD, CUPED) to measure impact on sales and guest counts for marketing decisions.
▸ Applied CUPED variance reduction (96%), uncovering statistically significant sales lifts of +$7,293 (DoorDash) and +$5,663 (Uber Eats) per store per week, informing channel prioritization and budget allocation.
▸ Developed interactive dashboards with parallel trends validation and regional breakdowns, enabling stakeholders to validate assumptions and analyze performance across stores.

Python Streamlit Causal Inference DiD CUPED

Snap Inc.

AR & ML Extern

▸ Engineered 5+ production AR Lenses in Lens Studio using computer vision and 3D rendering optimization, deployed across distributed cloud infrastructure serving millions of concurrent users at sub-50ms latency.
▸ Optimized rendering pipelines across 3 development cycles through systems-level performance profiling, improving frame rates and cutting memory footprint by 15%.
▸ Built A/B testing infrastructure with event tracking and distributed analytics pipelines, identifying UX bottlenecks and driving a 20% lift in key engagement metrics.

Lens Studio Computer Vision 3D Rendering A/B Testing

Aditya Birla Group

AI/Software Engineer Intern

▸ Built end-to-end distributed logistics optimization system using Python, scikit-learn, and TensorFlow, delivering $3.2M in annual cost savings with 23% delivery efficiency improvement across six enterprise clusters.
▸ Engineered a domain-specific Knowledge Graph in Neo4j modeling 500+ entities across 12 semantic dimensions, integrating Azure OpenAI with advanced prompt engineering to achieve 97% ontology accuracy on a production-grade LLM knowledge base.
▸ Developed automated document intelligence pipeline using fine-tuned transformer models and Hugging Face, achieving 96% extraction accuracy and reducing manual processing overhead by 80%.

Neo4j LangChain HuggingFace Azure OpenAI TensorFlow

Lawroom AI

Founding AI Engineer

▸ Architected and deployed a multilingual generative AI legal chatbot using InLegalBERT and ElevenLabs TTS, delivering accurate NLP responses across 10 languages to 2,000+ users.
▸ Designed systematic prompt engineering and LLM evaluation pipelines, improving translation reliability by 35% for low-resource Indian regional languages.
▸ Built scalable modular backend using FastAPI and Firebase with secure REST APIs, role-based authentication, and behavioral analytics supporting 3,000+ active users at production-grade uptime.

InLegalBERT FastAPI Firebase ElevenLabs TTS

Scale AI

AI Trainer, RLHF & Red Teaming

▸ Designed 500+ adversarial and edge case prompts to systematically expose failure modes in large language models including hallucination, instruction drift, and reasoning collapse across complex multi-turn conversations.
▸ Conducted red teaming evaluations across 3,000+ model responses, stress-testing alignment boundaries and surfacing vulnerabilities in LLM reasoning consistency.
▸ Maintained 90%+ agreement rate against ground truth labels, contributing high-quality feedback to RLHF pipelines used in production model training.

RLHF Red Teaming Hallucination Mitigation LLM Evaluation

Indian School of Business (ISB)

Research Intern

▸ Designed and deployed a production data engineering pipeline integrating PostgreSQL and MongoDB via REST APIs, improving data processing efficiency by 30% for research stakeholders.
▸ Automated metadata governance workflows using the OpenMetadata API, improving data discoverability and lineage tracking across the organization's data ecosystem.
▸ Built an interactive Streamlit data application delivering a self-serve interface for stakeholders to manage and explore organizational data assets independently.

PostgreSQL MongoDB Streamlit OpenMetadata

Omdena

Machine Learning Engineer

▸ Architected and deployed a generative AI mental health support chatbot using GPT-3 and transformer-based NLP, boosting user engagement by 40%.
▸ Fine-tuned NLP algorithms combining BERT embeddings and GPT-3 language generation, improving personalized advice accuracy by 30% for end users.
▸ Implemented a continuous feedback loop updating chatbot response quality from live user interaction data, improving conversational accuracy by 25%.

GPT-3 BERT NLP Python

Airtel

Machine Learning Intern

▸ Led optimization of a large-scale Python web scraping framework using multithreading and async I/O, reducing data extraction time by 30% across high-throughput production scraping jobs.
▸ Enhanced geospatial preprocessing pipelines using DBSCAN clustering and Folium, improving spatial accuracy by 20% for real estate property evaluation at scale.
▸ Designed a cross-validation algorithm reducing data discrepancies by 25% across multiple large-scale property data sources.

Python DBSCAN Folium Multithreading

Millennium Campus Network (MCN)

Campus Director, UN Millennium Fellowship

▸ Oversaw execution of a five-month UN Millennium Fellowship program, managing 30+ events and activities aligned with fellowship goals.
▸ Conducted bi-weekly progress reviews for 10 fellows, providing targeted feedback resulting in a 95% goal achievement rate.
▸ Served as primary liaison for 100+ stakeholders including fellows, mentors, and program administrators, fostering a high-collaboration environment across the full fellowship cohort.

Selected Work

Featured Projects

Production-grade AI systems and research-backed data solutions.

Agentic AI · Voice · Vision

OmniPro 220 Support Agent

Multimodal voice agent for the Vulcan OmniPro 220 welder built in 49 hours for Prox (YC F25). Hands-free voice loop, three-layer photo diagnosis with vision grounding, multi-agent deliberation (Safety + Quality reviewers), 14 structured tools, persistent cross-session memory, and customer mode toggle.

Next.js Claude API ElevenLabs TypeScript

Live Demo Code

Agentic AI · Automation

JobForge

Autonomous end-to-end job search pipeline. Scrapes Greenhouse, Lever, and Ashby APIs via Playwright, uses Claude Haiku for pre-filtering and Sonnet for deep evaluation, RAG over personal corpus via ChromaDB, auto-generates tailored CVs and cover letters, and submits applications autonomously. 97/97 tests passing.

Python Claude API ChromaDB Playwright

Code

CV · Generative AI · Research

InstantID — Facial Identity Persistence

Research project solving identity drift in diffusion models. Custom identity aggregation layer on top of InstantID using ArcFace embeddings. Aggregates 4-5 reference images into a single 512-dim master vector with confidence-weighted fusion. Targeting UCSD and Berkeley RA applications.

InsightFace ArcFace SDXL Gradio

Code

ML · Security

Real-time Fraud Detection

Ensemble ML model (XGBoost + Random Forest) for real-time credit card fraud detection achieving 99.2% accuracy, streamed via Apache Kafka and containerized with Docker.

XGBoost Random Forest Kafka Docker

Optimization · Supply Chain

Logistics Optimization Engine

Distributed route clustering system using ML and operations research that delivered $3.2M in annual cost savings and 23% faster delivery across six enterprise clusters at Aditya Birla.

Python scikit-learn TensorFlow PostgreSQL

LegalTech · Multilingual AI

Multilingual Legal AI Chatbot

Generative AI legal assistant using InLegalBERT + ElevenLabs TTS supporting 10 Indian languages, serving 3,000+ active users with production-grade availability at Lawroom AI.

InLegalBERT FastAPI Firebase ElevenLabs

View GitHub

Latest Writing

Insights

Writing on AI systems, technology, and the future of work. Published on Medium with 474+ reads on a single piece.

Mar 25, 2026 · 5 min read

Give Me 20 Minutes and I'll Change How You Use AI Forever

A practical reframe of how to get genuinely useful output from AI tools through structured interaction design.

Read on Medium →

Developer Tools

Mar 15, 2026 · 4 min read

The 5-Minute Setup That Turned My Terminal Into an AI Engineer

How a minimal terminal configuration unlocked a dramatically faster AI-assisted development workflow.

Read on Medium →

Future of Work

Mar 6, 2026 · 5 min read

The Hiring Pipeline Is Not Broken. It Is Being Quietly Restructured.

A clear-eyed look at what is actually changing in how companies hire AI and tech talent in 2026.

Read on Medium →

NLP Research

Nov 3, 2025 · 6 min read · 204 views

The Dash Addiction: Why AI Can't Stop Using Hyphens (And The Neural Math Behind It)

Exploring the structural reason LLMs over-rely on em-dashes, and what it reveals about token probability distributions.

Read on Medium →

Immigration · Tech

Sep 20, 2025 · 4 min read · 474 views

Why International Students Are Now the Cheapest Yet Priceless Talent in America

A data-backed analysis of the structural dynamics driving demand for international student talent in the US tech industry.

Read on Medium →

Agentic AI

Aug 10, 2025 · 9 min read · 143 views

How Graph Thinking Is Powering the $196B Agentic AI Revolution

Deep dive into knowledge graph architectures powering the next generation of agentic AI systems, and why GPT-5 changes the calculus.

Read on Medium →

Read All Articles on Medium

Recognition

Achievements

Scholarships, medals, and recognition across academics, research, and leadership.

🥈 National Competition

2nd Place, East Region at DataCamp's Data for Good Analytics Competition among 120+ graduate teams. Built an AI system to detect AI-generated misinformation in educational content, tackling factual reliability as AI becomes a primary learning tool.

🎓 Graduate Scholarship

Awarded 100% merit-based scholarship (Havells Foundation) to pursue MSBA at Wake Forest University, valued at $85,000+.

🏛️ Undergraduate Scholarship

Awarded the Bharati Scholarship, a full-ride covering four years of undergraduate education at Plaksha University ($42,000+).

🌐 International Recognition

Recognized by the National Academy of Engineering (USA) for research on entrepreneurial mindset in engineering education.

🥇 Case Study Champion

Gold Medalist at Caseify, a consulting-based case study competition on guestimation, defeating 12 competing teams.

⚙️ Engineering Innovation

Gold Medalist at PSIP and SP Dutta Innovation Award for designing a wind turbine with 8% higher output, selected from 40 competing teams.

🌍 Leadership

Selected as Campus Director for the UN Millennium Fellowship, managing 30+ events and leading 10 fellows to a 95% goal completion rate.

🏏 Sports

Intra-University Cricket Champion at Plaksha University, leading the team to the inter-department cricket tournament title.

✨ Spirit of Plaksha

Awarded the Spirit of Plaksha Award for outstanding contribution to peer learning, mentorship, and making technical concepts accessible to fellow students across the university community.

What People Say

Recommendations

From managers and colleagues who worked with me directly.

He took ownership of our website from design to Firebase integration, always with a user-centric focus. He also contributed to benchmarking AI models and integrating voice-to-text features. Vikas is constantly growing his skills and would be a valuable asset to any team.

Bhawna Rupani

Lead Gen AI Engineer · Lawroom AI

Vikas has consistently demonstrated a strong commitment to learning and delivering high-quality work on our Knowledge Graph implementation. He brings enthusiasm and curiosity to every task, communicates clearly, and collaborates effectively. His positive attitude and eagerness to learn make him a pleasure to work with.

Kunkum Poovamma

Senior GenAI Engineer · Aditya Birla Group

Vikas is a dedicated and hardworking individual with a strong passion for data science, machine learning, and deep learning. He consistently seeks opportunities to upgrade his skills and stays at the forefront of new technologies. His enthusiasm for ML is evident as he combines technical expertise with a genuine love for learning.

Tanisha Saraf

Robotics Engineer · Plaksha University

I build AI that
ships to production.

Real Impact. Real Production.

I treat prompt engineering as a systems design problem.

Education

Wake Forest University

Plaksha University

Work Experience

Wake Forest University

Darden

Snap Inc.

Aditya Birla Group

Lawroom AI

Scale AI

Indian School of Business (ISB)

Omdena

Airtel

Millennium Campus Network (MCN)

Featured Projects

Insights

Achievements

Recommendations

GitHub Activity

Let's Build Something

Send a Message

Nice to meet you

I build AI that ships to production.

Real Impact. Real Production.

I treat prompt engineering as a systems design problem.

Education

Wake Forest University

Plaksha University

Work Experience

Wake Forest University

Darden

Snap Inc.

Aditya Birla Group

Lawroom AI

Scale AI

Indian School of Business (ISB)

Omdena

Airtel

Millennium Campus Network (MCN)

Featured Projects

Insights

Achievements

Recommendations

GitHub Activity

Let's Build Something

Send a Message

I build AI that
ships to production.