LinkedIn Projects — Draft (8 total: 5 RBC + 3 Public R&D)

RBC / Current Role

Replaced a manual, weeks-long certification process (previously requiring multiple contract hires) with an AI-powered pipeline across 20+ device platforms — cutting execution time from weeks to hours and eliminating contractor dependency.
Built a PDF ingestion pipeline using prompt chaining + RAG to extract and normalize test specs from multi-page documents into structured, machine-readable markdown.
Automated test generation by dynamically scripting pytest/PyShark test cases and testbed YAML for scalable, repeatable validation.
Delivered centralized, audit-ready reporting (Allure-pytest) for full traceability — critical in a regulated, change-controlled environment.

Skills: LangChain, LangGraph, RAG, Prompt Engineering, pytest

Built a real-time topology and dependency-mapping platform giving ops teams blast-radius visibility before executing changes — directly reducing change-related downtime risk.
Designed Airflow DAGs for progressive, device-by-device data collection across multiple enterprise sources, avoiding load spikes at scale.
Modeled device relationships in Neo4j, exposing dependency chains and failure points instantly instead of via manual tracing.

Skills: Apache Airflow, Neo4j, Apache Kafka, Python

Architected a unified data platform pulling metrics from Postgres, Grafana, and Dynatrace into a single analytics layer — cutting data access latency 35% and giving stakeholders faster, proactive anomaly detection.
Built ETL pipelines (Kafka + Airflow) to extract, transform, and land metrics in AWS S3 for downstream analytics.
Delivered real-time monitoring dashboards replacing fragmented, tool-specific views.

Skills: ETL, Apache Kafka, Apache Airflow, AWS S3, Grafana

Delivered a natural-language chatbot letting engineers run multi-step network operations conversationally — replacing manual multi-command SSH workflows across Extrahop, MyOps, NetBrain, and Cisco ACI.
Refactored a monolithic architecture into Kafka-driven microservices, improving scalability and reliability under production load.
Added a MongoDB caching layer cutting redundant processing/query costs, plus WebSocket-based Webex/Slack notifications for live collaboration during changes.

Skills: LangChain, Generative AI, Apache Kafka, MongoDB, WebSocket

Extended an earlier natural-language ops chatbot into a LangGraph multi-agent architecture — a supervisor agent orchestrating domain-specific sub-agents for monitoring, troubleshooting, and automation — cutting manual troubleshooting time 40%.
Delivered a full-stack agentic solution (FastAPI, Kafka, MongoDB, React, TypeScript) with dynamic tool invocation and intelligent decision-making, extending beyond the original single-agent design.

Skills: LangGraph, FastAPI, React, TypeScript, Multi-Agent Systems

Benchmarked a fully local RAG pipeline (bi-encoder retrieval → CPU cross-encoder rerank → context compression → local LLM) achieving 948× token reduction (530K → 560 tokens) at 12/12 answer accuracy, with zero cloud API calls.
Compared FlashRank and BGE rerankers across Ollama-hosted Qwen/DeepSeek/GLM models to isolate which harness-level techniques actually move reliability — not just which model is "smarter."

Skills: RAG, LangGraph, Ollama, sentence-transformers, FlashRank

Designed a controlled, rung-by-rung benchmark isolating which harness components — planning, memory, verification, sandboxing, sub-agents — actually drive agent reliability, independent of the underlying model.
Ran three comparative experiments in LangGraph/LangChain against local Ollama models to produce reproducible, quantified reliability deltas rather than anecdotal claims.

Skills: LangGraph, LangChain, Ollama, Agent Evaluation

Built a high-throughput, serverless velocity/fraud-checking system for card transactions, modeling production-grade fintech architecture (API Gateway → SQS → Lambda → Redis) for real-time risk scoring.
Self-directed deep-dive into low-latency, event-driven fintech infrastructure patterns ahead of pursuing Capital Markets/Fintech roles.

Skills: AWS Lambda, Amazon SQS, API Gateway, Redis, Event-Driven Architecture