Prateek Deshmukh

Faff was a personal concierge on WhatsApp. I worked on most of it: agents, payments, internal tools, payouts, infrastructure and a fair amount of debugging.

• Shipped production consumer-facing AI agents
• Rebuilt unacknowledged message detection and cut latency from roughly two minutes to under 200 milliseconds.
• Built the internal system used to reconcile thousands of weekly vendor payouts.
• Moved payments from Razorpay to Cashfree and set up multilingual voice outreach.

Synapse routes a user’s request to the right workflow or specialist agent. I worked on the search and registry pieces behind that routing.

• Moved search from an in-memory vector store to PostgreSQL and pgvector.
• Added full-text search and BM25 ranking instead of relying only on embeddings.
• Connected Synapse to AWS Agent Registry and worked on the internal capability registry.

The fund had useful information spread across spreadsheets, LinkedIn, GitHub and personal sites. I built the pipeline that pulled it together, cleaned it up and found relationships between people.

• Ingested and enriched profiles from LinkedIn, GitHub and personal websites.
• Improved profile completeness by roughly 40–60%.
• Built a graph that could find the shortest introduction path between people.