AI Engineering · Agents
AI Agents Toolkit for Index Operations
Index operations are full of repetitive, error-prone steps: reading 100-page methodology PDFs, pulling Bloomberg data, and reconciling files from external calculation agents by hand.
- My role
- Designed the agent architecture, wrote 30+ of the 46 skills, and ran the rollout and training for both teams.
- When
- 2026 · ongoing
5 · 46
agents · skills
~30%
dev time saved (internal estimate)
40→6 min
to extract rules from a methodology PDF
8
analysts using it across 2 teams
Approach
Five specialised GitHub Copilot agents (rebalancing, code review, data, reporting, design) that orchestrate 46 small skills — each skill encodes one workflow with a structured output an analyst can check.
Results
The toolkit now handles the most repetitive parts of index operations. Rule extraction from a methodology PDF dropped from about 40 minutes to about 6, reconciliation reports come with a first-pass root-cause analysis, and internal estimates put the overall gain at roughly 30% of development time. The pattern was presented to a second trading team, which now uses it.
Architecture
Technical highlights
- 01Skills as small, composable units — one workflow each, testable in isolation
- 02Agents orchestrate skills; outputs are structured (Markdown specs, tables), never free text
- 03Human-in-the-loop by design: agents draft, analysts validate before anything reaches production
- 04Every skill ships with 3–5 golden examples used as regression tests
Stack
Related case studies
AI Engineering · Agents
Aloi, a Market Research Agent that Checks its Figures
27/28
eval questions passed with no figure missing from the fetched data (one run)
AI Engineering · Developer Tools
AgentDeck, One Place to Answer Your Coding Agents
1 place
to answer Claude Code and Codex, on the PC or from a phone
AI Engineering · RAG
Index Methodology RAG Assistant
81.9%
accuracy on 94 held-out questions (95% CI 73–88%)