Agentic AI
Supervisor-Agent AI System
+400
hours/year of first-time-right production
Shubham Jain · Forward Deployed Engineer
Builds systems
people use every day.
Agentic AI & prescriptive ML for manufacturing.
I embed with plant operators, scope the problem on the ground, and ship what they end up using every day: supervisor-agent systems with RAG and memory, multi-horizon forecasting with optimization, and conversational data apps running inside a legacy alumina plant.
Selected work.
Systems shipped to production, measured by what changed on the ground.
Simulation & optimization
Calcination Digital Twin
Daily
used by plant operators
Fraud detection
AI Claims Engine
40%
less manual processing · 18% accuracy uplift
Spatiotemporal ML
Disease Hotspot Forecasting
₹5M+
business value generated
At a glance.
More projects, where I've worked, and what I work with. Tap any tile for details.
Databricks App
Conversational Plant Data App
Self-serve
process data, operator to leadership
Forecasting & optimization
SSA Forecast & Prescribe Engine
↑ FTR
raised first-time-right rates
Experience
Embedded where the work happens.
Trainee→Data Scientist→FDE
- Aditya Birla Management CorporationForward Deployed EngineerJan 2025 – Present
- Fractal AnalyticsBusiness Analyst InternMay 2024 – Jul 2024
- Subhkam VenturesQuant Analyst InternJun 2023 – Jul 2023
Agentic AI
NL2SQL Agent
1st
GDNA Hackathon 2026 winner
Quant research
Quant Trading Platform
+15%
vs NIFTY50 · 25% return improvement
Stack
Agents, forecasting and optimization, shipped on Databricks.
Multi-Agent OrchestrationRAGLangGraphLightGBMXGBoostDoubleMLGA / PSODigital TwinsPythonPySparkDatabricksAzure DevOps
Education · 2021 – 2025
B.Tech, Data Science, Economics & Business
Plaksha University
9.0 / 10 CGPA
Certifications
- Databricks GenAI Engineer Associate (2026)
- Databricks ML Associate
- GDNA Hackathon 2026 Winner
- VMart Scholarship — Top 1%
- Toastmasters Speechcrafter (2025)
Open to opportunities
Let's build something real.
Interested in roles where I can solve complex business challenges and create measurable impact.
Agentic AI
Supervisor-Agent AI System
+400
hours/year of first-time-right production
One agent that routes every plant question to the right tool, and remembers the answer.
Supervisor-agent architecture with intent classification and tool routing across RAG (Databricks Vector Search), Genie Spaces, and model endpoints, backed by session and long-term memory on Lakebase. Replaced manual analyst intervention in plant decision workflows.
A middle manager for AI agents, except it actually delegates well. Analysts got their afternoons back.
RAGVector SearchGenie SpacesLakebase memory
Simulation & optimization
Calcination Digital Twin
Daily
used by plant operators
A virtual calciner that tries every trade-off first, so the real plant doesn't have to.
Analytics layer of a calcination digital twin, built end to end: high-frequency sensor data fused with ML models for simulation-based optimization across production, quality, and energy trade-offs.
Moved into an alumina plant for weeks, learned to speak fluent operator, then built the plant a helpful twin.
Sensor fusionML modelsSimulationMulti-objective
Fraud detection
AI Claims Engine
40%
less manual processing · 18% accuracy uplift
Flags suspicious insurance claims before an investigator has to dig.
LLM-SLM-XGBoost hybrid for insurance claims fraud detection, designed, trained, and deployed end to end. Language models handle document understanding and feed an XGBoost fraud classifier.
Taught AI to read insurance claims like a skeptical detective.
Fraud detectionDocument understandingEnd-to-end deployment
Spatiotemporal ML
Disease Hotspot Forecasting
₹5M+
business value generated
Multi-week risk forecasts that flag outbreaks before they happen.
Spatiotemporal ML model for early disease hotspot detection, built on geospatial and time-series feature engineering, delivering multi-week risk forecasts for proactive underwriting.
A crystal ball for disease outbreaks. No magic involved, just spatiotemporal ML.
GeospatialTime seriesFeature engineering
Databricks App
Conversational Plant Data App
Self-serve
process data, operator to leadership
Operators ask in plain English; leadership gets its own view.
Databricks App (Streamlit/React) for conversational querying of process data, with role-based operator and leadership views, a KPI hierarchy (FTR, throughput), optimization prescriptions, alerting, and data-quality pages.
Plant data, but you can just talk to it. Nobody has to pretend they understand the other's dashboard anymore.
Databricks AppsStreamlitReactKPI hierarchy
Forecasting & optimization
SSA Forecast & Prescribe Engine
↑ FTR
raised first-time-right rates
Forecasts, evolutionary search and causal ML, with a prescription-vs-actual scorecard.
Multi-horizon forecasting and prescriptive optimization for Specific Surface Area (SSA) in calcination: LightGBM forecasts paired with GA/PSO search and DoubleML causal estimates, with prescription-vs-actual tracking.
Forecasting, plus a swarm of particles, plus a little Darwinian evolution, with causal ML as the designated adult.
LightGBMGA / PSODoubleML
Agentic AI
NL2SQL Agent
1st
GDNA Hackathon 2026 winner
Business teams query data conversationally, without SQL.
Natural-language-to-SQL agent that lets business teams query data conversationally, without writing SQL or waiting on an analyst.
SQL is a love language most business teams don't speak. So I built an agent that translates.
LLM agentSQL
Quant research
Quant Trading Platform
+15%
vs NIFTY50 · 25% return improvement
Momentum strategies across 1,100+ equities.
Proprietary momentum-based trading algorithms with a systematic backtesting framework, plus portfolio optimization and risk management across a large equity universe.
Math beats gut feeling every time.
MomentumBacktestingPortfolio optimization
Experience
Where I've worked.
Aditya Birla Management Corporation
Forward Deployed Engineer · Jan 2026 – Present
- Embedded with plant operators for weeks to scope the problem on the ground, then led the analytics layer of a calcination digital twin end to end across production, quality, and energy trade-offs; operators now use it daily.
- Architected and deployed a supervisor-agent system with intent classification and tool routing across RAG, Genie Spaces, and model endpoints, with session and long-term memory on Lakebase; adds 400 hours/year of first-time-right production.
- Shipped a Databricks App for conversational querying of process data with role-based views, a KPI hierarchy, optimization prescriptions, alerting, and data-quality pages.
- Sole owner of the stakeholder chain from plant operator to CEO, including adoption after go-live.
Management Trainee, Data Scientist · Jul 2025 – Jan 2026
- Built a multi-horizon forecasting and prescriptive optimization engine for SSA in calcination (LightGBM + GA/PSO + DoubleML) with prescription-vs-actual tracking.
- Shipped an LLM-SLM-XGBoost hybrid for claims fraud detection: 18% accuracy uplift, 40% less manual processing time.
- Built a spatiotemporal disease-hotspot model delivering multi-week risk forecasts worth ₹5M+ in business value.
Data & Analytics Trainee · Jan 2025 – Jun 2025
- Fine-tuned a small language model on plant terminology for low-latency, domain-specific operations queries.
- Built an OPC UA connector streaming live sensor data from 4+ lines into PostgreSQL via Databricks/PySpark, with CI on Azure DevOps.
Fractal Analytics
Business Analyst Intern
- Translated B2B campaign data into predictive and clustering models with client stakeholders, driving a 10% lift in lead conversion for a high-value cohort.
- Designed and ran A/B tests to isolate high-impact customer segments across multiple touchpoints.
Subhkam Ventures
Quant Analyst Intern
- Formulated quantitative strategies using moving averages and momentum indicators, achieving 15% outperformance against NIFTY50.
- Deployed optimization models that raised returns by 25% on a diversified portfolio of 1,100+ cash stocks.
Stack
Full tech stack.
GenAI & Agentic AI
Supervisor/Multi-Agent OrchestrationIntent Classification & Tool RoutingRAGDatabricks Genie SpacesSession/Long-Term Memory (Lakebase)SLM Fine-TuningPrompt EngineeringLLM EvaluationLangChainLangGraphSemantic Search
Machine Learning
Multi-Horizon ForecastingLightGBMXGBoostSpatiotemporal ModelingFeature EngineeringCausal Inference (DoubleML)SHAPA/B TestingMLflowChampion/Challenger
Optimization & Simulation
Genetic AlgorithmsParticle Swarm OptimizationPrescriptive AnalyticsDigital TwinsConstraint Optimization
Languages & Data
PythonPandasNumPyScikit-learnPyTorchSQLPySparkPostgreSQLOPC UA
Platforms & Delivery
DatabricksDatabricks Apps (Streamlit/React)Azure DevOpsGitAWSPower BI