Read app screenshots for 69% less than GPT-6.1 Sol, at 97% of its field accuracy

Build the "Screenshot mining" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: google/gemma-4-31B-it (OpenAI-compatible endpoint: /v1/chat/completions). Keep the model id configurable.

Task
- Input: an uploaded app screenshot plus the fields to extract.
- Behaviour: return every field in the submitted schema as validated JSON
- Send one POST /v1/chat/completions request (SDK: chat_completions / chatCompletions) with a system message and a user message. The system message is "Read only the supplied full screenshot and return one JSON object conforming to the schema. Do not use external knowledge or invent hidden content. Copy strings as visibly displayed, preserving spelling, case, diacritics, signs, currency symbols and punctuation. Values that are absent, blank, only placeholders or not fully readable are null. Checkbox states are booleans; counts are integers; money and dates are strings. Include every complete requested row in visible order. Ignore navigation and fields not requested. Do not output Markdown, commentary or reasoning.\n\nExtract the displayed task reference, title, labelled summary fields, dates and description. Copy the main task values only; exclude action links such as Assign to me. Use null for a requested time-spent value when that line is absent. Include every complete subtask row in its top-to-bottom order, with its displayed time-tracking string.\n", a blank line, "JSON schema:" and the submitted schema as JSON indented by two spaces; response_format only constrains decoding, so the model reads the field descriptions from this text. The user message holds only an image_url part. Base64-encode the original local image bytes as a canonical data:image/png, data:image/jpeg or data:image/webp URI (normalize jpg to image/jpeg). Use max_completion_tokens 32768, temperature 0 and response_format {type: "json_schema", json_schema: {name: "fixture_kb01", strict: true, schema}} with the submitted schema unchanged. Parse choices[0].message.content as JSON and fail loudly when it is missing or invalid. Submitted schema: {
  "type": "object",
  "properties": {
    "task_reference": {
      "type": [
        "string",
        "null"
      ]
    },
    "title": {
      "type": [
        "string",
        "null"
      ]
    },
    "status": {
      "type": [
        "string",
        "null"
      ]
    },
    "priority": {
      "type": [
        "integer",
        "null"
      ]
    },
    "complexity": {
      "type": [
        "integer",
        "null"
      ]
    },
    "category": {
      "type": [
        "string",
        "null"
      ]
    },
    "swimlane": {
      "type": [
        "string",
        "null"
      ]
    },
    "column": {
      "type": [
        "string",
        "null"
      ]
    },
    "position": {
      "type": [
        "integer",
        "null"
      ]
    },
    "assignee": {
      "type": [
        "string",
        "null"
      ]
    },
    "creator": {
      "type": [
        "string",
        "null"
      ]
    },
    "time_estimated": {
      "type": [
        "string",
        "null"
      ]
    },
    "time_spent": {
      "type": [
        "string",
        "null"
      ]
    },
    "due_date": {
      "type": [
        "string",
        "null"
      ]
    },
    "created": {
      "type": [
        "string",
        "null"
      ]
    },
    "modified": {
      "type": [
        "string",
        "null"
      ]
    },
    "description": {
      "type": [
        "string",
        "null"
      ]
    },
    "subtasks": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "title": {
            "type": [
              "string",
              "null"
            ]
          },
          "assignee": {
            "type": [
              "string",
              "null"
            ]
          },
          "time_tracking": {
            "type": [
              "string",
              "null"
            ]
          }
        },
        "required": [
          "title",
          "assignee",
          "time_tracking"
        ],
        "additionalProperties": false
      }
    }
  },
  "required": [
    "task_reference",
    "title",
    "status",
    "priority",
    "complexity",
    "category",
    "swimlane",
    "column",
    "position",
    "assignee",
    "creator",
    "time_estimated",
    "time_spent",
    "due_date",
    "created",
    "modified",
    "description",
    "subtasks"
  ],
  "additionalProperties": false
}. Preserve top_k when supplied, presence_penalty 0 and the submitted whitespace setting.

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.