I work at the intersection of digital lending, credit risk, AI/ML, product engineering, enterprise architecture, and operational transformation.
My focus is not model experimentation for its own sake. I work on AI systems that can operate inside regulated financial environments: explainable, observable, auditable, measurable, and useful to business teams.
I think in three layers:
| Layer | What it means |
|---|---|
| Business layer | Lending journeys, risk policy, underwriting, operations, compliance, conversion, and customer experience |
| AI/ML layer | Document intelligence, bureau parsing, feature engineering, decisioning, RAG, local LLMs, evaluation, drift, explainability |
| Engineering layer | APIs, microservices, orchestration, observability, dashboards, CI/CD, secure deployment, and production support |
|
|
|
| Area | Tools and direction |
|---|---|
| AI/ML | Python, scikit-learn, PyTorch, NLP, GenAI, RAG, model evaluation, drift analysis |
| LLM systems | LangChain, LangGraph, Ollama, local LLMs, vector search, prompt engineering, agent orchestration |
| Backend | FastAPI, Flask, Node.js, REST APIs, microservices, async workflows |
| Frontend | React, Next.js, Flutter, dashboards, workflow UIs, product prototypes |
| Data | PostgreSQL, MongoDB, Redis, vector databases, analytics pipelines |
| Cloud and DevOps | AWS, Docker, Kubernetes, CI/CD, GitHub Actions, monitoring, secure deployment |
These are connector-verified lower-bound indicators from accessible repositories, not decorative widget totals.
| Signal | Verified evidence |
|---|---|
| Recent authored PR records surfaced | 200 PR records surfaced across two connector queries |
| TradeOn PR depth | TradeOn PR sequence reaches #249 |
| Saarthi PR depth | saarthi-ai-agent PR sequence reaches #93 |
| Recent authored commit records surfaced | 100 latest commit records surfaced in one connector query |
| Activity pattern | Frequent PRs across trading systems, AI agents, LLM observability, CI/CD, iOS builds, deployment fixes, diagnostics, and production hardening |
Short summaries only. No core code, proprietary diagrams, schemas, prompts, internal workflow logic, API contracts, credentials, deployment details, or confidential assumptions are exposed.
| Private / concept work | Short public-safe description |
|---|---|
| Saarthi | AI-assisted guidance layer for complex financial-service journeys |
| Journey Bot | Conversational support for assisted digital workflows |
| Merchant Bot | Merchant-facing AI assistant for sales, inventory, offers, and service workflows |
| TradeOn | Market workflow and trading-system exploration with risk-first thinking |
| Lensr | LLM evaluation, response inspection, and observability harness direction |
| Agentic AI | Agent orchestration, memory, approval controls, and enterprise automation patterns |
| FlowForge | Enterprise AI-native workflow and low-code builder platform direction |
| Cadence Flow | Enterprise workflow, productivity, and project intelligence platform direction |
These repositories are public for portfolio review and professional evaluation. They are not open source. Each repo uses a Proprietary / All Rights Reserved license, and reuse requires written permission.
| Repository | Short description | License type |
|---|---|---|
| UPI FlowPilot | UPI checkout reliability engine with flow selection, retry intelligence, degradation detection, mock NPCI/UPI responses, and merchant SRE workflows | Proprietary / All Rights Reserved |
| Bharat UPI Interdict | Pre-settlement mule-network interdiction and fraud investigation copilot with graph-risk scoring, hold simulation, and explainable fraud reason codes | Proprietary / All Rights Reserved |
| Cashflow Memory for Bharat | Explainable UPI and Account Aggregator-style cashflow memory engine for thin-file merchant credit-readiness and responsible lending workflows | Proprietary / All Rights Reserved |
| UPI Guardian Mode | AI consent firewall for agentic UPI payments with bounded authority, prompt-injection risk checks, step-up decisions, and human-review controls | Proprietary / All Rights Reserved |
| UPI Social Proof Ledger | AI verification layer for payment claims, fake UPI screenshots, QR/VPA mismatch, seller trust, and dispute-ready social-commerce proof | Proprietary / All Rights Reserved |
| UPI Cognitive Spend Brake | Responsible digital-spending friction layer for UPI payment intents, impulse-risk detection, UPI Lite leakage, budget nudges, and self-control rules | Proprietary / All Rights Reserved |
AI in financial services fails when it remains a model demo.
It works when it is wrapped with governance, explainability, audit trails, monitoring, exception handling, human override, security, and business ownership.
The model is only one component. The real system is the operating layer around the model.
Applied AI for financial services. Built with product thinking, engineering discipline, and risk awareness.