Skip to content
Portrait of Bhavya Verma

Bhavya Verma

AI Engineering & Data Science · Novartis Risk & Compliance

I help teams turn manual workflows into AI systems they can trust.

Hyderabad, India · Open to work worldwide

Selected work

Novartis

Natural-language analytics agent

Agentic text-to-SQL assistant that lets business users query governed compliance data in plain English, with per-user access guardrails.

Internal

Novartis

Automated audit control testing

LLM pipeline that tests ~700 anti-corruption audit questions across 30 controls. Cited verdicts, benchmarked against human reviewers.

Internal

Novartis

Global-to-local config sync

LLM matcher with a verifier loop and accuracy evals that syncs global workflow changes to 77 country configs. ~90% less manual effort.

Internal

Novartis

RAG framework for compliance assistants

Tool-calling RAG framework with hybrid search and feedback telemetry, behind three deployed compliance assistants.

Internal

BlackCoffer · 2023

Route optimisation for field crews

OR-Tools route planning for installation crews. 25% less travel, 30% better on-time performance.

Code on GitHub for Route optimisation for field crews

BlackCoffer · 2022–2023

Energy price forecasting

Fuel-price forecasting (LSTM / GRU) on a GCP data pipeline. 50% faster data processing.

Code on GitHub for Energy price forecasting
More work (7)

  • Audit-intelligence dashboard · Novartis

    Clusters audit observations by meaning and over time to surface recurring themes, then writes an executive briefing grounded in computed metrics.

  • Forecasting HSE metrics with Chronos · Novartis

    Benchmarked a time-series foundation model against statistical baselines; forecast MAPE from 139% to 25%.

  • Policy consolidation engine · Novartis

    Finds overlapping and subset policy documents with retrieval and chunk-coverage analysis to support keep / drop / merge decisions.

  • Record linkage for missing reporting IDs · Novartis

    Fuzzy-matching candidate generation followed by LLM adjudication between vendor and CRM records.

  • Snake species detection · IIIT Delhi

    CNN image classifier (DenseNet, Inception V3, ResNet) to help identify snakes in snake-bite treatment. GitHub for Snake species detection

  • Smart sign-language gloves · IIIT Delhi

    Real-time American Sign Language-to-speech with wearable gloves, tested with users from several NGOs.

  • Speech recognition for regional languages · IIIT Delhi

    Course project combining deep neural networks and hidden Markov models. GitHub for Speech recognition for regional languages

Experience

  1. Jul 2024 – present

    Data Analytics and Insights Specialist, Risk & Compliance · Novartis

    15+ GenAI and ML applications shipped to production.

  2. Aug 2022 – Apr 2024

    Data Scientist · BlackCoffer

  3. Oct 2021 – Jul 2022

    Freelance Data Scientist

  4. Jun 2021 – Sep 2021

    Software Engineering Intern · Autodesk

  5. 2017 – 2021

    B.Tech, Electronics & Communication Engineering · IIIT Delhi

AI
Agents & tool calling · RAG · LLM evals · Guardrails · Human-in-the-loop · Document extraction · Observability
ML & data
Anomaly & drift detection · Entity resolution · Forecasting · Python · SQL · PyTorch · Snowflake · Databricks · AWS · Azure · Docker · Kubernetes

Work with me

Open to AI engineer and data science roles anywhere, and to focused projects on LLM features that need to be reliable.

bhavyaverma1699@gmail.com

© 2026 Bhavya VermaViews are my own, not Novartis’s.