Synechron —
Selected projects

Real-time AI platform simulating live customer calls for training call center agents. Agents speak to AI-powered personas — frustrated, calm, confused — and receive instant performance feedback across five dimensions: empathy, problem-solving, professionalism, de-escalation, and follow-through.

• Reduced end-to-end response latency from ~15s to sub-3s via async processing and WebSocket streaming.
• Built a 5-dimension LLM evaluation framework with custom prompt engineering and agent scoring.
• Containerized with Docker; CI/CD via AWS CodePipeline + CodeBuild.

Modular AWS-based platform for rapidly deploying AI solutions for enterprise clients. Designed around a multi-account architecture (Core, LOB, CI/CD) with centralized authentication and environment provisioning.

• Documented Cognito authentication design across account tiers — Core requires Cognito; LOB and CI/CD use cross-account IAM roles.
• Contributed to infrastructure provisioning templates and Active Directory integration for client onboarding.

ML recommendation system matching investors to alternative investment deals — real estate, private equity — based on risk profile, income, investment horizon, and behavioral signals. Built two recommendation pipelines plus a secondary model predicting investment likelihood.

• Trained KMeans clustering to segment 607 investors into 5 risk tiers with continuous scores from centroid distances.
• Built content-based filtering via deal similarity and collaborative filtering via matrix factorization on a 265×26 interaction matrix.
• Handled cold-start with demographic-based user similarity for investors with no history.
• Feature engineering from sparse data — 5,597 rows with up to 94% missingness on key columns.