LedgerSight
Shipped a production indexer + natural-language query interface in 8 weeks; analyst query time dropped from 30 minutes to 30 seconds.
Client / Project
LedgerSight is a mid-tier DeFi protocol running on Ethereum + a major L2. Their growth team needed real-time visibility into pool flows, liquidation events, and arbitrage activity — but the only path to answers ran through a backend engineer running custom RPC queries by hand.
Problem
The protocol's existing analytics ran on a stale 24-hour ETL job. Every time the growth team wanted a fresh slice — "which liquidations cascaded today?", "what's the TVL trend in the last 4 hours?" — an engineer had to write a fresh query against the node. Decisions slowed to the engineer's calendar.
Approach
A 3-engineer Project Pod: one Web3 engineer for the indexer, one backend engineer for the API + agent layer, one full-stack for the analyst UI. We indexed the relevant contracts into Postgres with a lightweight schema (events + derived state) and exposed it through an agent that translated natural-language questions into SQL behind the scenes.
The agent layer used Claude with a tightly-scoped tool: read-only SQL against the indexer schema, with row caps and timeout guards. The analyst could ask "show me liquidations over $50k in the last hour" and get a chart in seconds — no engineer in the loop.
Outcome
Analyst query time dropped from "ping the on-call engineer, wait 30 minutes" to "ask the agent, get an answer in 30 seconds." Indexer p95 latency stayed under 2 seconds even during high-volume liquidation events. The growth team started running 8x more queries per week, and the on-call engineer reclaimed roughly a day of attention each week.