Projects

Financial analysts were checking 50+ data sources manually ” public markets, premium feeds, internal trade logs. Missing signals. Drowning in noise.

Built an AI system that watches everything. Discover and analyze data daily. Surfaces only what matters.

What it does:

• Synthesizes multiple data streams in real-time (premium analyst content, market news feeds, public filings, internal credit analysis)
• AI reasoning identifies material risks and opportunities based on analyst preferences
• Automatic prioritization by urgency and impact
• Processes 10K+ events daily ” alerts on the top 5. Configurable by user.

The mainframe was a 40-year-old beast. Processing billions in transactions. Couldn’t just flip a switch.

Migrated it piece by piece over 18 months. Business kept running.

How it worked:

• Strangler pattern migration from Db2 on z/OS to PostgreSQL ” zero data loss
• Decomposed monolith into event-driven microservices using Kafka for async communication
• Containerized services on Kubernetes with auto-scaling and self-healing
• 60% infrastructure cost reduction while improving performance and developer velocity

Executives needed threat visibility. Security team had mountains of data. No way to connect them.

Built a mobile-first dashboard. Real-time alerts. Drill down from summary to raw events.

The stack:

• Data pipeline using Spark and AWS Glue processing millions of security events daily
• Native iOS app with WebSocket feeds for instant threat notifications
• Interactive drill-down from executive summary to individual security events
• Unified API layer serving both mobile and web clients

Investment team had data scattered across 12 systems. Near-real-time price, purchase, & disposal feeds. Daily batch reports. No single source of truth.

Built a unified warehouse with tiered pipelines. Different data, different speeds.

The architecture:

• Near real-time pipelines for volatile market data (prices that change by the second)
• Daily batch for static datasets (fundamentals, earnings, holdings)
• Snowflake-based warehouse with curated data products for different teams
• RESTful APIs exposing investment metrics and portfolio analytics
• Model Context Protocol (MCP) integration ” LLMs query investment data directly

Employees spent hours hunting for policies. Asking managers. Pinging coworkers on Teams. “Where’s the PTO policy?” “How do I expense this?” Constant frustration.

Built a chatbot with RAG. Get the answer in seconds instead of asking around.

How it works:

• Vector-based semantic search using pgvector for document retrieval
• Responses grounded in company documentation (no hallucinations)
• Plain English questions, specific answers
• Policy lookups went from hours to seconds ” employees love it