MY JOURNEY

Experience

From classical NLP and scientific computing to speech AI and intelligent agents, my experience spans different areas of software and AI engineering.

Here is a closer look at the problems I have worked on, the approaches I explored, and the engineering skills I developed along the way.

Sep 2026 – Present · Remote

ANCI AI

AI Engineer Intern

Improving the reliability, security, and usability of an AI-powered scheduling platform.

Designed 293 security test cases across 3 priority tiers (222 critical, 50 high-priority, 21 hardening) to systematically evaluate authorization, user/team isolation, and scheduling workflow risks.
Executed and documented security tests across authenticated and unauthenticated scenarios, comparing expected and observed behavior to identify access-control and data-isolation defects.
Developed multi-turn scheduling scenarios involving time zones, external participants, approvals, conflicting availability, and email interactions to evaluate agent behavior beyond simple successful bookings.
Documented reproducible failures, partial outcomes, and unresolved cases with supporting evidence and follow-up actions, helping engineering teams prioritize investigation and remediation.
Contributed to scheduling UI improvements, persona-driven email-response testing, and product recommendations while coordinating engineering work through Jira.

AI AgentsSecurity TestingAPI TestingWorkflow TestingJira

May 2025 – Aug 2025 · Noida, India

Data Science Intern

Investigating and optimizing Hindi text-to-speech inference using F5-TTS.

Adapted the open-source F5-TTS inference pipeline for Hindi by integrating a Hindi-trained checkpoint and vocabulary with the F5TTS_Small architecture.
Profiled inference performance and traced computational overhead to the conditional flow-matching pipeline; evaluated local asset loading and caching, identifying why caching did not improve generation for changing input text.
Built a Hindi speech-data preparation pipeline covering audio extraction, enhancement, speaker cleanup, silence trimming, segmentation, normalization, Whisper transcription, and language filtering.
Adapted the fine-tuning workflow for F5TTS_Small, prepared training artifacts, and ran fine-tuning experiments with loss tracking through Weights & Biases.
Experimented with reducing ODE inference steps from 32 to 8, evaluating the trade-off between generation speed and speech quality; further reductions noticeably degraded quality.

PythonPyTorchF5-TTSWhisperWeights & BiasesAudio Processing

Feb 2024 – Jul 2024 · Bengaluru, India

Data Analytics & AI Intern

Analyzing regional sales drivers and improving regression-based modeling for an FMCG client.

Analyzed SKU-level sales drivers across markets in Africa, the Middle East, and South Asia, investigating how pricing, promotions, weather, holidays, competition, and product cannibalization influenced demand.
Adapted variable selection to individual markets after identifying that no single predictor combination consistently explained sales behavior across regions.
Evaluated regression models using R², adjusted R², MAPE, elasticity, and statistical significance, refining predictor sets to balance model performance with interpretability.
Investigated missing competitor and cannibalization data, handled outliers, and applied base-price smoothing to improve data quality before model evaluation.
Incorporated region-specific demand events, including local festivals, holidays, and major cricket matches, when evaluating potential sales drivers.
Prepared SKU-level Excel reports with model results and visualizations, supporting analysis, review, and discussion with the project team.

Regression AnalysisFeature SelectionStatistical ModelingExcelData Cleaning

Mar 2023 – Jul 2024 · Tsukuba, Japan (Remote)

Distributed Computing R&D Intern

Improving error diagnostics and functional testing for distributed scientific computing tools.

Contributed to the Belle II experiment's Grid Computing (gb2) toolset, used to support distributed computing workflows across collaborating research institutions.
Implemented standardized error-code handling across gb2 tools, replacing ambiguous failure states with more structured and diagnosable errors.
Designed and executed disk-quota boundary experiments to examine system behavior before and after storage thresholds were reached.
Built a pytest-based functional test suite for the gb2_job toolset to validate expected behavior and improve repeatability of testing.

PythonpytestGrid ComputingError HandlingJira

Jun 2022 – Aug 2022 · India

NLP & Machine Learning Intern

Developing multilingual language identification and interpretable document topic-modeling pipelines.

Developed multilingual language-identification experiments using 2 vectorization methods (CountVectorizer and TF-IDF) and 2 classifiers (Multinomial Naive Bayes and SVM).
Applied preprocessing, train/validation/test splitting, k-fold cross-validation, and grid-search hyperparameter tuning to compare classification performance.
Investigated classification difficulties between related languages and tested resampling, class priors, class weighting, and sample weighting to address class imbalance.
Reported F1 scores of 0.98 for weighted Multinomial Naive Bayes with CountVectorizer and 0.99 for weighted SVM with TF-IDF in the internship experiments.
Developed an Anchored CorEx topic-modeling workflow to organize documents into interpretable topics, exploring hierarchical representations and total correlation.
Experimented with domain-guided topic discovery using anchor words and anchor strengths, documenting a configuration with 5 anchor words and an anchor strength of 5.

Pythonscikit-learnNLPSVMNaive BayesTF-IDFCorEx