Software Engineer
V2 Solutions Inc. · Jul 2024 – Present
- Developed, deployed and led 2 interns in building a LangGraph based Multi-Agent Fashion Trends aggregator and analyzer for Perspiq.ai, an apparel enrichment platform, with two specialized agents: a Trend Aggregation Agent that runs hourly to collect fashion signals from web sources, generate vector embeddings, and persist them in pgVector; and a Trend Recommendation Agent triggered daily to match client catalogs against emerging trends using Cosine Similarity and Deterministic Matching and deliver actionable recommendations via email, including the relevant product, target market/area, and expected trend duration. improving inventory utilization of our clients by 30%
- Built a LangFuse LLM-as-Judge evaluation suite scoring the Trend Recommendation Agent's outputs against a fashion-domain rubric (product-trend fit, market relevance, duration accuracy), flagging 15% of runs as low-confidence for prompt and retrieval tuning, improving judge/human agreement from 78% to 91% over three iterations
- Collaborated directly with Finance and Accounts teams to develop an Agentic AR Automation System using AWS Agentcore, Bedrock & Strands API, fully automating the processing of remittance emails, payment-to-invoice reconciliation, and ERP updates reducing the daily reconciliation cycle by almost 99.98%
- Implemented persistent RAG-based agent memory to checkpoint and retrieve insights from previous runs, enabling contextual continuity and progressively improving agent decision-making.
- Engineered reliable and secure agent execution using AgentCore Gateway for ERP integration, Bedrock Guardrails for PII/Financial Data redaction, and HITL workflows to route ambiguous payments for human validation.



