Bengaluru, India · Open to ML/AI roles

I build ML systems that survive contact with real data.

Machine Learning Engineer at IISc working across forecasting, MLOps, computer vision, and agentic AI. I take systems from messy inputs through evaluation, deployment, monitoring, and a usable interface.

Explore my work View resume

profile.yaml

role: ML Engineer
focus:
  - time-series forecasting
  - production MLOps
  - applied GenAI
strength: end-to-end ownership
currently: IISc Bengaluru
preferred: applied ML in a domain

3,552ward-month records modeled

11.84%sMAPE on AquaOps

6tool-using agents in Sutra

80%exact result match in DataLens

TIME-SERIES FORECASTING◆PRODUCTION MLOPS◆ APPLIED GENAI◆COMPUTER VISION◆ GEOSPATIAL ANALYTICS◆FULL-STACK DELIVERY◆ TIME-SERIES FORECASTING◆PRODUCTION MLOPS◆ APPLIED GENAI◆COMPUTER VISION◆ GEOSPATIAL ANALYTICS◆FULL-STACK DELIVERY◆

Current work

Applied ML for urban water systems

Indian Institute of Science 2024 — Present

Groundwater Outlook for Bengaluru

I solely own the ML workflow developed in collaboration with BWSSB: reconciling erratic measurements, engineering temporal features, selecting models through leakage-safe backtests, and communicating operational limits.

37Bengaluru wards

222final-test predictions

0.606aggregate final-test R²

0.907R² at shorter horizons

Read the methodology ↗

Acquire

CGWB manual, DWLR, and state groundwater records.

Reconcile

Create a complete monthly target series without target imputation.

Validate

Use fixed-origin backtests and an untouched temporal test.

Explain

Report horizon degradation and operational assumptions honestly.

Selected builds

Systems, not notebook demos

01 / MLOps↗

AquaOps

Production-style urban water-demand forecasting with controlled promotion, drift monitoring, and conditional retraining.

7,271 monthly records across 44 zones
Beat the best baseline in 3 of 4 windows
MLflow · DVC · PSI drift · FastAPI · Docker

Champion model11.84% sMAPE

02 / Agentic AI↗

Sutra

A deployed multi-agent chief of staff with real tools, streaming traces, memory, and human confirmation before outbound actions.

FastAPI + React + TypeScript
Calendar and Gmail OAuth integrations
Dockerized on Google Cloud Run

Architecture5 agents + learner

03 / Text-to-SQL↗

DataLens

Natural-language database analytics with hybrid schema retrieval, constrained SQL generation, validation, and correction retries.

FAISS + BM25 with rank fusion
Read-only policy enforced by sqlglot
EXPLAIN before query execution

Benchmark80% exact match

04 / RAG↗

PaperLens

Citation-aware PDF question answering using dense and sparse retrieval, rank fusion, and local cross-encoder reranking.

Page-level source metadata
Answer-linked citation filtering
Up to 5 PDFs and 200 pages

RetrievalHybrid + reranked

05 / Applied AI↗

MatchLens

Multi-facet resume matching with structured skill gaps and longitudinal job-market drift analysis.

Four weighted embedding facets
Centroid distance and KS testing
SQLite + Parquet persistence

FocusInterpretable matching

06 / Google ADK↗

SmartText Agent

Containerized tool-calling agent for summarization, Q&A, classification, text statistics, and intent routing.

Five deterministic tool contracts
Offline tests and GitHub Actions
Deployed to Google Cloud Run

DeliveryTested + deployed

Trajectory

From learning models to owning systems

The through-line is increasing ownership: first models, then research workflows, then deployed and monitored systems.

2022

Foundation

ML Intern

TensorFlow training, inference refinement, and client-pipeline integration.

2024

Applied research

IISc Project Intern

Environmental forecasting, data reconciliation, and computer vision research.

2025

Ownership

IISc Project Assistant

Sole ownership of a city-scale forecasting workflow with stakeholder delivery.

2026

Production thinking

MLOps + Agentic AI

Deployment, monitoring, controlled promotion, human-in-the-loop systems, and CI.

How I work

Engineering judgment before model novelty

I am most useful where data is inconsistent, model behavior needs to be defended, and the final system has to work for someone outside the ML team.

My work spans environmental forecasting, MLOps, computer vision, RAG, text-to-SQL, full-stack deployment, and stakeholder communication.

Validate against reality

Temporal splits, untouched tests, baselines, and explicit failure analysis.

Prefer evidence over complexity

A simpler model wins when it generalizes better and costs less to operate.

Build the whole path

Data, model, API, UI, deployment, monitoring, and documentation belong together.

PythonScikit-learnTensorFlowLightGBM FastAPIDockerGCPMLflow DVCGeminiLangChainFAISS SQLReactGIS

Let’s build something useful

Looking for an ML engineer who can own more than the model?

I’m targeting ML Engineer, AI Engineer, and Applied Scientist roles across India.

Email me LinkedIn GitHub