Arjun Bindu Jayachandran

About

Ask my portfolio

A miniature of the agent architecture I build at work: an orchestrator routes your question, a retriever pulls grounded context, and a validator checks the answer before it renders. Scripted demo — answers are precomputed, no live LLM, nothing leaves your browser.

Experience

1. Jan 2026 – Present

   Senior Software Engineer

   Wells Fargo · Bengaluru

   • Building a machine-learning bond recommendation engine that surfaces the most relevant bonds for each institutional client from a daily sheet of 500–800 offerings. Own the learning-to-rank models (XGBoost), the feature pipeline, and an LLM assistant traders use to explore similar bonds and give feedback that improves future recommendations. Rankings are validated in offline backtesting on historical trades.
   • Led a performance investigation into slow PnL calculations, benchmarking each service route to isolate the bottleneck. Reworked request throttling and load distribution in the calculation path, cutting end-to-end p95 latency by 66%.
   • Shipped historical rerun support for the PnL service using async Camel routes over MongoDB and Kafka. Corrections to past positions recompute and persist under the correct historical date without republishing into the live trade stream.
   • Championed GitHub Copilot adoption across the line of business, presenting to 200+ engineers on practical usage and agentic development workflows.
2. Jul 2022 – Dec 2025

   Software Engineer

   Wells Fargo · Bengaluru

   • Designed and prototyped a natural-language interface to MongoDB that turns plain questions into safe queries. Orchestrator and validator LLM agents (LangGraph, tool access via MCP) run retrieval-augmented generation over schema and user-context vector stores, behind guardrails that mask PII and block unsafe queries.
   • Rebuilt the real-time PnL read path on MongoDB, replacing view-based reads with aggregation pipelines and compound indexes. P95 read latency dropped from around 710ms to 230ms with no loss of accuracy.
   • Extended the Java Spring Boot PnL microservices to a new asset class: expected cash-flow models, EUR FX conversion, swap-position roll-ups, and risk-factor attribution. Released with production Kafka topics and regression sign-off.
   • Created a reusable activity-tracking layer adopted by two trading-desk teams across six service flows and eight repositories. It emits run status with correlation IDs, and its non-blocking design keeps monitoring off the critical path.
   • Converted a PnL backup service from per-record calls to batched writes, raising end-of-day throughput by roughly 90% and easing load on shared infrastructure.
   • Automated regression testing across 150+ financial scenarios with parallel execution and Splunk telemetry, reducing run time by 60%, and drove the migration of 10+ services onto OpenShift with rebuilt CI/CD pipelines.
   • Delivered an asset-backed-finance reporting feature end to end on an investment-banking platform: an ingestion engine for Excel deal and pricing sheets, plus the schema migrations and reference-data reconciliation behind it.
3. May 2021 – Jul 2021

   Technology Intern

   Wells Fargo · Bengaluru

   • Prototyped a time-series forecasting engine for ATM cash-replenishment optimization, benchmarking SARIMA, LSTM, and Prophet across 111 ATMs with holiday and weather signals. Reached about 91% accuracy on weekly forecasts.

Selected work

Personal builds along the same lines as my Wells Fargo work — agentic LLM pipelines, retrieval, and applied ML — plus tooling I needed for myself. Plenty more side experiments on GitHub.

Publications & Awards

Skills

Education