From clinical NLP to industrial AI — each stop added a layer: data engineering, deployment, then grounded GenAI.
AIA Engineering Ltd — Ahmedabad
End-to-end AI/ML pipelines for predictive maintenance, demand forecasting, and process optimization; real-time analytics architectures; model governance and continuous-learning pipelines for Industry 4.0 manufacturing.
Sambhav Foundation — Bengaluru
Scalable Python/SQL/Pandas ETL over multi-state beneficiary data (+40% data availability); Flask KPI dashboards driving +35% decision quality; validation automation that cut reconciliation effort by 40%; mentored junior analysts.
Access Automation Pvt. Ltd. — Bengaluru
92%-accuracy sentiment model; Flask anomaly dashboards (-45% response time); CV-screening engine automating 75% of resume filtering; churn prediction with Naïve Bayes + LSTM (+28% retention).
AI Variant — Bengaluru
Medical data preprocessing (+33% intake accuracy); clinical NLP feature engineering; TensorFlow cardiovascular-disease models at 94% accuracy, +12% over baseline.
From notebook to production — grounded LLM apps, air-gapped analytics, and the ML pipelines behind them. Tap any project to dive in.
LangChain · Vector DB · RAG
LangChain + FAISS/ChromaDB retrieval over OpenAI & Gemini for grounded enterprise Q&A with citations or explicit abstention.
What's in flight right now — present tense.
Shipping predictive-maintenance & demand-forecasting pipelines for Industry 4.0.
Hardening Polaris — the air-gapped, LLM-never-does-arithmetic data analyst.
Owning the roadmap for an internal AI copilot — writing PRDs, scoping the eval harness, and running the discovery-to-launch loop.
Standing up a metrics layer so every AI feature ships with the activation, retention, and cost dashboards to prove it out.
Designing a personal 3D portfolio around the Rahu / eclipse metaphor.
From raw ETL to a deployed model behind an API — I own the whole path: data engineering, feature work, model development and evaluation, then productionize with Flask/FastAPI/Streamlit.
Best for: teams that need one person to take AI from notebook to production.
Grounded GenAI that cites its sources or abstains — retrieval pipelines, document intelligence, and agentic Plan→Tool→Execute→Reflect loops with human gates. Target: fully grounded answers, zero unsupported claims.
Best for: enterprises that need trustworthy, citable answers over their docs.
Enterprise-safe AI that runs entirely on-prem — no cloud dependency at runtime, designed within real hardware ceilings (CPU-only, 16 GB RAM) with graceful deterministic fallbacks.
Best for: regulated or offline environments where data can't leave the building.
I turn data into decisions — pipelines, models, and RAG agents that ship.
Real outcomes and the recognition that followed. References to collect below.
Certificate of Excellence for leading an AI product from discovery to launch — recognized for innovation and measurable business impact.
Cut planning-cycle time 60% by shipping an ML scheduling optimizer — reframed as a product with clear success metrics, not a one-off model.
Drove adoption of an internal AI copilot to 80% weekly-active across the team by owning the roadmap, PRDs, and rollout.
Essays and posts on RAG, agents, MLOps and where deep learning is heading — on Medium and LinkedIn.
MediumAug 17, 2026
LinkedInJul 30, 2026
LinkedInJun 23, 2026
LinkedInJun 20, 2026
LinkedInMay 30, 2025
MediumApr 29, 2025
MediumApr 26, 2025
MediumApr 24, 2025
LinkedInSep 27, 2024
LinkedInSep 19, 2024End-to-end AI systems: data pipelines, ML/DL models, and RAG/agentic LLM apps — from ETL to a deployed, monitored product.
Both, but I default to local-first / air-gapped for anything sensitive. When cloud APIs are used, there's an explicit boundary with redaction/opt-out.
Grounding, citations, and abstention over false confidence — and a hard rule that LLMs never generate numbers that flow into outputs; deterministic code does the math.
Python/FastAPI/LangChain on the backend; Next.js/React/TypeScript/Tailwind (and Three.js/R3F for 3D) on the front; vLLM/Ollama + Qwen for local models.
Based in Ahmedabad, India, and open to AI/ML engineering roles and select consulting — remote or on-site. The fastest way to reach me is email.
Let's build something grounded
I take AI from notebook to production — grounded, local-first, and reproducible. Tell me what you're building.