Gnaneshwar Vanam
7 years building production AI/ML, GenAI, and AI Agent solutions delivering measurable business impact across pharma, healthcare, and enterprise environments. Full-stack AI engineer (Next.js · TypeScript · FastAPI · Python). Claude Certified Architect.
About Me
Senior Data Scientist with 7 years building production AI/ML systems and deploying production-grade ML, GenAI, and LLM solutions across pharma, healthcare, and enterprise environments — from $10M+ in delivered savings to 10,000+ manual hours automated annually.
My stack: Python · TypeScript · PyTorch · LangChain · FastAPI · Next.js · Azure. I own the full pipeline — feature engineering, model training, REST API deployment, monitoring, and automated retraining.
Recent focus: agentic systems with Claude & Gemini, RAG architectures over large document corpora, LoRA/QLoRA fine-tuning for domain-specific tasks, and MLOps pipelines on Azure that stay reliable under production load.
Core Stack
Experience
Senior AI Engineer / Forward Deployed
GenAI Data Quality Agent
Scoped, designed, and productionised a multi-turn Gemini 3 agentic system extracting structured fields from 10,000+ unstructured promotional campaign texts. Hit 92% field-level accuracy with 2% validation errors using a hand-labelled eval set, a two-layer validator with self-correction, tool calls, and retry loop — automating ~10,000 analyst hours annually.
Pixel Dynamic Pricing Engine
Led end-to-end build of a LightGBM regression and constrained-optimisation engine on EMEA market signals. Shipped an interactive dashboard for analysts to visualise forecasts, adjust constraints, and re-run optimisation on demand, replacing week-long feedback cycles — delivering 90% forecast accuracy and $10M incremental profit in 6 months.
Senior AI Engineer / Forward Deployed
Fine-Tuned LLM + RAG for Sales Analytics
Owned discovery, scoping, and production deployment of a domain-specific RAG system backed by a fine-tuned LLM (LoRA / QLoRA). Built an automated eval harness scoring answers against a labelled golden dataset. Hit 95% answer accuracy and reduced analyst effort by 40%.
Customer Churn Intelligence
Built an XGBoost churn prediction model with 80% precision, identifying key churn drivers with SHAP-based explainability. Monitored production drift and enabled 95% retention of high-priority B2B customers, protecting $10M+ in annual revenue.
AI Fraud Detection
Engineered supervised risk-scoring models combining account-compromise signals and free-tier behavioural patterns. Achieved 92% precision and mitigated ~$4M in potential revenue loss.
Data Scientist / AI Engineer
Hospitalisation Risk Prediction
Developed a predictive model to identify patients at high risk of unplanned inpatient hospital admissions, enabling proactive intervention and care planning — improving precision by 20% and generating $15M in savings.
Cohort Analysis & Drug Adoption
Led a 5-member data science team using clustering and predictive modelling to boost drug adoption by 5%. Applied Shapley values for explainability, driving stakeholder confidence in model decisions.
NLP & Admission Prediction
Applied BERT vectorisation, PCA, and cosine similarity for patient cohort analysis. Built and deployed ML models predicting inpatient admissions with 3% precision improvement.
Model Productionisation & MLOps
Productionised multiple predictive models with robust data pipelines, monitoring, and retraining workflows on Azure.
Personal Projects
AI/ML systems I've designed, built, and shipped — from GenAI agents to full-stack AI applications.
GiftVision
AI Product Customisation Platform
Chat with Gemini 3.0 Flash to customise awards, trophies, and gifts via natural language — no editing software required. Describe your vision and the AI instantly generates a realistic product image, ready to go straight to hardware production.
Users select a base product from the gallery, then describe desired changes in plain English — 'Add Taylor as Employee of the Year', 'Happy Valentine's Day with a red ribbon', 'Happy Birthday Sarah'. Gemini 3.0 Flash receives both the base product image (multimodal input) and the user prompt, generating a modified product image instantly. Outputs go straight to hardware production — no designer or editing tools needed.
- ▸No editing software required — describe in plain English
- ▸AI-generated images ready for hardware production
- ▸Conversational chat interface with full message history
- ▸Wide variety of gift types: trophies, awards, plaques & more
- ▸Multimodal AI (image + text input)
- ▸Real-time image generation
Truth Verifier
AI-Powered Health Misinformation Detector
Paste an Instagram post or reel URL and get real-time fact-check verdicts — TRUE, FALSE, or UNVERIFIABLE — with cited sources in seconds. Built to make health and science fact-checking frictionless at scale.
The backend fetches Instagram content via yt-dlp (cookie-based auth), extracts audio with ffmpeg, and transcribes it via Gradium STT. TinyFish Agent AI parses the transcript into individual factual claims, searches each against the web using TinyFish Search API, then returns a verdict with explanation and source citation. Results stream to the browser in real time via SSE — users see claims resolve as they arrive, no full-pipeline wait.
- ▸Real-time SSE streaming — verdicts appear as each claim resolves
- ▸Solved Vercel Edge timeout by moving DB write to client-side POST after 'done' event
- ▸Tiered usage system (free / pro / admin) with live usage banner and upgrade modal
- ▸File-based extraction cache keyed by URL + vision flag — avoids redundant Gradium calls
- ▸Supabase RLS with service-role bypass pattern for reliable server-side writes
- ▸Vision mode: Gemini Vision analyzes on-screen text in video frames (admin tier)
More Projects
Campaign Intelligence Platform
GenAI Agentic Data Extraction System
Multi-turn GenAI agentic loop that extracts structured fields from unstructured telecom promotional campaign text — partner, product, promotion type, and offer details — with 90%+ accuracy.
The system runs a multi-turn LLM conversation with tool use: database lookup for partner resolution, two-layer business rule validation, and automatic self-correction retry. Supports OpenAI GPT-4o and Google Gemini interchangeably. Full audit trail logged as structured JSONL.
- ▸Multi-turn agentic loop (up to 10 turns)
- ▸Two-layer validation + self-correction
- ▸Provider-agnostic (OpenAI / Gemini)
- ▸Full JSONL audit trail
Hackathon — Audience Builder
Fractal Internal Hackathon · 1st Place
Built an Audience Builder tool that applies ML to identify the right customer and right product for upsell based on existing product usage patterns, achieving 85% precision. Surfaces productivity metrics and audience segments to drive data-backed commercial decisions. Presented to Chief Practice Officer at Fractal AI.
End-to-end ML pipeline: ingests existing product usage data, applies clustering and classification to surface customer-product upsell pairs, and visualises audience segments and productivity metrics in AWS QuickSight. SageMaker pipeline handles model training and serving. Presented to Chief Practice Officer at Fractal internal hackathon.
- ▸ML-powered customer + product matching for upsell
- ▸Productivity metrics surfaced for commercial teams
- ▸Audience segmentation by product usage patterns
- ▸Presented directly to Chief Practice Officer
Agentic Job Discovery Engine
Autonomous Career Page Scraper
Agentic tool that reads companies and job keywords from a Google Sheet, searches for career pages via Google Search + Gemini disambiguation, scrapes listings with Playwright, and writes results back automatically.
End-to-end autonomous pipeline: reads from Google Sheets → searches for career pages → scrapes job listings with headless Chromium → uses Gemini to disambiguate results → writes structured output back to the sheet. Handles pagination, dynamic JS-rendered pages, and multi-tab layouts.
- ▸Headless browser scraping
- ▸Gemini-powered page disambiguation
- ▸Google Sheets read/write
- ▸Handles dynamic JS-rendered pages
Certifications
Architect
Claude Certified
Architect
Anthropic · April 2026
Certified in designing, building, and deploying production-grade AI systems using Claude models — covering advanced prompt engineering, multi-agent architectures, tool use, safety practices, and enterprise AI integration patterns.
Additional Certifications
Awards & Recognition
Technical Skills
Languages & Data
Frontend
ML & Deep Learning
Generative AI & LLMs
MLOps & Cloud
Backend & Infra
Core Skills Radar
Tools & Frameworks
Get In Touch
Let's Build Something Impactful
I'm actively exploring senior AI/ML and Data Science opportunities where I can drive meaningful impact through GenAI products, ML systems, or strategic data roadmaps.
Whether you're building a new AI product, scaling an ML platform, or planning your data strategy — let's connect.
OPEN TO OPPORTUNITIES
Ready to Connect?
Send me an email and let's discuss how I can contribute to your team's success.
Gnaneshwar Vanam
Senior Data Scientist · AI/ML Engineer · GenAI Architect · London