I'm an Applied AI Engineer and Portfolio Analytics professional at Shriram Finance.
I enjoy understanding how people work, identifying recurring operational bottlenecks, and building reliable systems that eliminate them.
Most of the systems I've built started as real problems I encountered at work, repetitive reporting, manual document processing, underwriting workflows, and analytical requests that consumed valuable time.
My engineering approach is simple:
- Understand the workflow.
- Find the recurring friction.
- Design a reliable solution.
- Use AI only where it genuinely adds value.
I'm currently building AI-assisted systems for lending, portfolio analytics, and workflow automation.
Open to Applied AI Engineer, AI Engineer, GenAI Engineer, Analytics Engineer, and Data Analyst opportunities.
Technology should follow the problem, not the other way around.
I believe deterministic systems should handle anything requiring correctness, while AI should be responsible for language understanding and reasoning, not inventing business metrics.
Across almost everything I build, the goal is the same:
Reduce the time between information and business decisions by removing recurring operational bottlenecks.
🛠️ Tech Stack
Collection leaders repeatedly depended on analysts for portfolio summaries, KPI tracking, branch comparisons, and ad hoc analytical requests. Preparing those insights often took hours.
CollectionIQ is a portfolio intelligence platform that enables business users to explore lending portfolios using natural language.
Instead of relying on an LLM to calculate business metrics, it separates language understanding from numerical computation. AI interprets user intent, while deterministic analytics compute every KPI through validated business logic.
- Used daily by business leaders
- Reduced portfolio insight turnaround from hours to under one minute
- Eliminated repetitive analytical requests
- Enabled self-service portfolio exploration using natural language
Python • LangGraph • Pandas • Gemini API • Streamlit • LangSmith
Underwriters spend significant time validating borrower information, reviewing bureau reports, and preparing credit assessments.
FinSight combines deterministic underwriting rules with AI-assisted reasoning. Rule-based validations ensure correctness, while AI provides contextual risk summaries and grounded borrower Q&A.
- Automated 15 underwriting validations
- Reduced manual underwriting review effort
- Improved consistency of credit assessment workflows
- Human-in-the-loop decision support
Python • LangChain • Gemini API • PyMuPDF • Streamlit
Loan officers manually extracted hundreds of loan accounts from bureau reports before underwriting could even begin.
AutoCAM automates bureau extraction, validates outputs against bureau summaries, and uses AI only when deterministic extraction methods require assistance.
- Reduced processing time from 30–60 minutes to under one minute
- Supports multiple bureau formats
- Validation-first architecture for reliable extraction
- AI fallback only when required
Python • PyMuPDF • Gemini API • OpenPyXL • Streamlit
- Observe the workflow.
- Identify recurring friction and operational/infomational bottleneck.
- Design a deterministic solution.
- Use AI only where reasoning adds value.
- Measure success by time saved, reliable system and better decisions.
- CollectionIQ v2
- Enterprise Underwriting Intelligence
- AI Evaluation & Reliability
- Workflow Automation for Lending Operations
