Varun Viswanath

Lead ML Engineer

Mumbai, Maharashtra, India2 yrs 9 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Designed AI voice agents for enterprise workflows.
Developed automated grievance classification for Indian Railways.
Co-created regression testing framework for ML notebooks.

Stackforce AI infers this person is a Machine Learning Engineer with expertise in AI-driven solutions across various sectors.

Contact

Skills

Core Skills

Artificial Intelligence (ai)Large Language Models (llm)Machine LearningNatural Language Processing (nlp)Test AutomationFull-stack DevelopmentData Science

Other Skills

Shell ScriptingGenerative AIPython (Programming Language)Data EngineeringDeep LearningRegression TestingTransformersNLTKKerasReact.jsNode.jsStatistical Data AnalysisChatbot DevelopmentWeb ScrapingAmazon Web Services (AWS)

About

Agent Engineer at Prodigal with a background in applied machine learning and NLP. Previously built and deployed ML systems for Indian Railways (RailMadad) and contributed to research on regression testing for ML notebooks (UIUC Research). Interested in building reliable, scalable ML systems, especially in the NLP and LLM space.

Experience

2 yrs 9 mos

Total Experience

1 yr

Average Tenure

11 mos

Current Experience

Prodigal

Agent Engineer

Jun 2025 – Present · 11 mos

● Designing and deploying production AI voice agents for enterprise debt collection workflows, handling 1000+ daily interactions with high reliability across multi-step state-driven conversations (verification, negotiation, payments)
● Building tool-augmented LLM systems, integrating backend services (CRM, payments, verification APIs) to enable agents to take real actions beyond conversation
● Architecting multi-tenant backend systems for AI agents:
> extensible tenant client frameworks
> config-driven field mappings
> reusable tool interfaces across clients
● Designed state-aware agent workflows, ensuring consistent transitions across conversation stages and handling real-world edge cases (timeouts, partial failures, API inconsistencies)
● Built and optimized data pipelines and sync systems (Databricks → PostgreSQL), including:
> tenant-triggered sync architecture
> concurrency-safe UUID-based table swaps
● Improved backend performance and query efficiency:
> implemented suffix-based search across multiple fields
> achieved ~77% latency reduction in PostgreSQL queries
● Developed evaluation and analytics frameworks to monitor agent performance (transfer rates, call outcomes, latency breakdowns)

Artificial Intelligence (AI)Shell ScriptingGenerative AILarge Language Models (LLM)Python (Programming Language)Data Engineering

Centre for railway information systems (cris)

Machine Learning Engineer

Dec 2024 – Jun 2025 · 6 mos · Remote

Worked on an automated grievance classification system for RailMadad, the national complaint portal of Indian Railways (10,000+ daily submissions), replacing manual type/subtype selection with real-time ML inference.
● Designed and deployed a MaskedDense Tri-LSTM architecture enforcing logical type–subtype consistency (94.9% type accuracy, 89.6% subtype accuracy) with millisecond-level inference latency
● Integrated the model into the live RailMadad workflow, enabling automatic classification directly from grievance text input
● Built a Markov chain-based synthetic data generation pipeline to create 35,000+ complaints across 59 underrepresented categories, filling 88% of sparse entries and improving dataset coverage by 4.7%
● Developed a label validation pipeline using zero-shot NLI models (SummaCZS, SummaCConv) to detect mislabels, improving effective subtype accuracy by 18.7%
● Contributed to a production ML system serving national-scale public grievance resolution

Artificial Intelligence (AI)Deep LearningMachine LearningNatural Language Processing (NLP)

University of illinois urbana-champaign

Research Intern

Jun 2024 – Jun 2025 · 1 yr · Remote

Under the guidance of Dr. Saikat Dutta, co-developed NBTest, an automated regression testing framework for ML Jupyter notebooks.
● Designed a custom pytest plugin that auto-generates cell-level assertions by analyzing dataset characteristics, model configurations, and evaluation metrics
● Used AST parsing to extract ML pipeline structure and identify testable properties within notebook cells
● Introduced statistical tolerance bounds for stochastic ML metrics, reducing test flakiness to <1%
● Evaluated across 194 Kaggle notebooks (~33 assertions per notebook)
● Developed ML-specific mutation operators and validated effectiveness on historical buggy notebook versions, detecting 83% of pipeline bugs (e.g., data leakage, hyperparameter misconfigurations)

Python (Programming Language)Test AutomationRegression TestingMachine Learning

Piramal capital & housing finance limited

Business Intelligence Unit Intern

Jun 2024 – Aug 2024 · 2 mos · On-site

● Automated CIBIL Report generation using an optimized LLM-driven pipeline, reducing generation time by 20% and maintaining accuracy with JSON-based context processing.
● Enhanced HR AI Chatbot by integrating advanced conversational memory, improving context retention and accuracy in chatbot responses.
● Optimized Hallucination Detection for RAG by using open source Lynx model, cutting response time by 50%+ while maintaining reasoning accuracy.

Artificial Intelligence (AI)Large Language Models (LLM)Data Engineering

Simppl

Responsible Computing Fellowship

Jan 2024 – Jul 2024 · 6 mos · Hybrid

Synapse

2 roles

Machine Learning Head

Promoted

Aug 2023 – Jul 2024 · 11 mos

TransformersArtificial Intelligence (AI)Deep LearningNLTK

ML Developer

Sep 2022 – Aug 2023 · 11 mos

Natural Language Processing (NLP)Python (Programming Language)Keras

Dj unicode

2 roles

Fullstack Node Mentor

Aug 2023 – Jun 2024 · 10 mos

Full Stack Developer

Oct 2022 – Aug 2023 · 10 mos

Full-Stack DevelopmentReact.jsNode.js

Infiheal

Data Science Intern

Jun 2023 – Aug 2023 · 2 mos · Remote

● Developed interactive mental health AI using Langchain and GPT-3.5 for conversational user interaction.
● Performed data engineering and cleaning on Reddit chats and synthetic data to build datasets for sensitive topic classification.

Data ScienceStatistical Data AnalysisMachine LearningChatbot DevelopmentWeb ScrapingAmazon Web Services (AWS)+1