Hello — I’m Vishnu


building production GenAI.

Lead ML Engineer (8 years). Diffusion pipelines, multi-provider LLM orchestration, MCP servers, and the MLOps to keep them all up at 3 AM.

Portrait of Vishnu Anilkumar
scroll ↓
02

About & Resume

  1. Lead Machine Learning EngineerGenAI SaaS (marketing platform)

    Nov 2022Present

    Own the AI Python backend across two production GenAI products — multi-modal creative generation and a workspace-scoped LLM analytics service with an MCP surface.

    • Drove a creative-generation platform from inception to $2M+ ARR; led a 4-person AI team through v1 launch
    • Shipped two production MCP servers (11+ tools combined)
    • 25+ Baseten-hosted models in production, multi-provider LLM orchestration
  2. Senior Machine Learning EngineerGlobal AI Consultancy

    May 2021Nov 2022

    Led CLTV engagements for Fortune-500 retail and consumer-goods clients and mentored a team of seven MLEs. Recognised as Employee of the Year.

    • Production CLTV pipeline on Kubeflow + GCP AI Platform
    • North-star metric ownership across data, modeling, and sign-off
  3. Machine Learning EngineerProduct Engineering Firm

    Apr 2020May 2021

    Skill-extraction CNN, BERT semantic clause search, YOLOv3 signature detection.

    • 82% NER accuracy
    • BERT + Elasticsearch search engine
  4. Associate Data ScientistData Science Consultancy

    Jun 2018Mar 2020

    Resume intelligence, churn prediction, cattle face-identification CV.

    • 82% churn-prediction accuracy on 22K records
  5. Intern Data ScientistData Science Consultancy

    Dec 2017Jun 2018

    Probability-of-default modelling on KYC data — multiple model comparison, class-imbalance handling, anomaly detection.

    • SMOTE-balanced PoD models
  6. M.Sc. Computer Science (Machine Intelligence)IIITM-K (CUSAT), Kerala, India

    20162018

    Specialisation in Machine Intelligence.

  7. Employee of the YearGlobal AI Consultancy

    Recognised for flawless delivery and leading the ML team across multiple Fortune-500 engagements.

Featured stack

Python95%
PyTorch90%
Stable Diffusion XL90%
MCP (FastMCP)90%
FastAPI95%
AWS (EKS/S3/CloudFront)85%
03

Selected Projects

  • GenAI2023–Present

    Multi-Modal Creative GenAI Platform

    46-route production platform — diffusion, video, brand QA — inception to $2M+ ARR.

    PythonFastAPISQLAlchemy (async)Celery+11
  • GenAI2024–Present

    Unified Ad Generation Pipeline

    One-call orchestrator — 4 generation modes, 12 downstream stages, one schema.

    FastAPICeleryRabbitMQBaseten+7
  • LLM Agents2024

    Multi-Provider Ad Copy Engine

    Headlines, captions, CTAs — brand-voice-grounded, platform-aware, campaign-scale.

    OpenAI GPT-4oAnthropic ClaudeGoogle GeminiFastAPI+2
  • LLM Agents2024

    Prompt Enhancer & Variation Engine

    Turn a 5-word brief into 12 structured variants, each tuned for a different generation stage.

    OpenAIClaudeFastAPISemantic diversity scoring+2
  • GenAI2023–2024

    AI Image Editor Suite

    Inpainting, outpainting, object removal, reference-guided enhancement — 7 tools, one editor.

    LaMaSAM (Segment Anything)SDXL inpaintFlux+3
  • GenAI2024

    Product-to-Video Studio

    Static product shot in → cinematic motion out, with camera choreography and continuity checks.

    Diffusion video modelsVision LLMffmpegBaseten+2
  • GenAI2024

    Talking-Avatar UGC Video Pipeline

    Studio-free UGC ads: avatar + voice-clone + camera-movement, batched at fleet scale.

    DiffusionElevenLabsVision LLMffmpeg+4
  • LLM Agents2024–Present

    Ad Intelligence MCP Server

    Workspace-scoped ad-library analytics with an 11-tool MCP for Claude Desktop & Cursor.

    PythonFastAPIFastMCP 3.1SQLAlchemy 2.0 (async)+10
  • LLM Agents2024–Present

    Creative QA MCP Server

    The second MCP in the stack — brand-compliance QA and creative heatmaps, tool-callable.

    PythonFastMCPAnthropic Claude (vision)GPT-4o+4
  • LLM Agents2023

    RAG Pandas Agent

    Conversational data ops for non-technical users — plain English in, dataframes out.

    LangChainChromaPandasOpenAI
  • MLOps2022

    Enterprise Retail CLTV Pipeline

    Ensemble Customer-Lifetime-Value model on GCP AI Platform with production monitoring.

    KubeflowGCP AI PlatformBigQueryPython
  • CV/NLP2020–2021

    Skill-Extraction CNN

    Entity recognition for an enterprise hiring platform — resumes to skill graphs.

    PyTorchCNNNERFlask
  • CV/NLP2020–2021

    Legal-Clause Semantic Search

    BERT + Elasticsearch search engine for fast clause comparison across large legal corpora.

    BERTElasticsearchFlaskPython
  • CV/NLP2020–2021

    Signature & Stamp-Seal Detection

    YOLOv3 computer-vision model automating proofreading of legal documents.

    YOLOv3OpenCVPyTorchFlask
  • MLOps2021–2022

    Dawn Foods — Customer Lifetime Value

    CLTV with monthly revenue as the North Star — owned end-to-end as Senior MLE.

    PythonBigQueryGCPPandas+1
  • MLOps2021–2022

    Henry Schein — Next-Purchase Forecasting

    Predictive reorder model for sales operations on a B2B medical-distribution catalogue.

    PythonPandasscikit-learnXGBoost+1
  • MLOps2022

    Bank of Montreal — NBP & CLTV Architecture

    Pre-sales solution architecture for Next-Best-Product and Customer-Lifetime-Value modelling.

    GCP AI PlatformBigQuerySolution architectureStakeholder presentation
  • Classical ML2018–2020

    Customer-Churn Prediction

    82% accuracy on 22K customer records for a financial-services client — XGBoost + targeted retention.

    XGBoostPandasscikit-learnAWS Elastic Beanstalk
  • CV/NLP2018–2020

    Resume Intelligence Platform

    Candidate scoring · section ID · content-quality eval · alt-industry/role prediction.

    TensorFlowRandom ForestBERT (fine-tuned)Pandas+1
  • CV/NLP2019–2020

    Cattle Face-Identification

    Computer-vision claim verification for India's largest insurer — automated cattle insurance.

    OpenCVCNNPythonPaperspace
  • Classical ML2017–2018

    Credit-Risk Analytics — Probability of Default

    KYC-driven probability-of-default modelling with SMOTE, anomaly detection, and feature selection.

    RegressionSVMRandom ForestXGBoost+2
04

Skills & Tools

22 tools8 at lead83% avg

Hover or tap a chip to see how I’ve used it.

06

Say hello