Output Parsers

Output Parsers in InnoSynth-Forjinn enable AI workflows to convert raw LLM or agent output into structured, validated, and machine-usable formats. They ensure downstream nodes, actions, or integrations can reliably consume model responses.

---

Why Use Output Parsers?

• Structured Data: Convert model outputs to JSON, objects, or domain-specific types.
• Robustness: Prevents breaking changes if LLM output format changes; you enforce what’s expected.
• Validation: Reject or repair malformed model outputs.
• Integration: Feed output to APIs, databases, or other automation confidently.

---

Usage in Workflows

• Add Output Parser node downstream from LLM, agent, or any Tool node
• Choose expected structure/schema (e.g., JSON, CSV, object field mapping)
• Connect to downstream nodes for API requests, storage, or further processing

---

Types of Parsers

1. JSON Parser

• Tries to extract/validate a JSON object from text output or responses
• Able to auto-fix minor format issues (common with LLMs)

2. CSV/TSV Parser

• For tabular outputs, parses LLM/txt output to arrays/records

3. Custom Schema/Object

• Define expected keys/types; parser validates output and provides error if format is invalid

4. Domain Parsers

• Special plugins for e.g., SQL, YAML, Markdown tables, XML

---

Configuration

• Input: Connect from the node producing unstructured output (LLM, Tool)
• Schema (if supported): Optional; define keys, types, required/optional status
• Error Handling: Set fallback value, retry, or alternative route when parsing fails

---

Example Use Case

1. LLM Node generates Q&A responses expected as { "question": "...", "answer": "..." }
2. Output Parser Node: set schema to enforce keys/types
3. If valid: route to storage/API; if invalid: log error or prompt user to try again

---

Troubleshooting

• LLM returns invalid JSON: Add system prompt instructing proper format; use Output Parser "auto-fix" when available
• Fields missing: Set schema keys as optional or configure fallback path

---

Best Practices

• Use Output Parsers when integrating with any system that requires structured data (webhook, CRM, DB)
• Combine with Moderation for safe, sanitized, AND structured output
• For complex tasks, break up output into simpler key/value pairs

---

Output Parsers are your safeguard against LLM/AI unpredictability—ensure every integration always gets clean, actionable data.