diff --git a/src/gemini_wrapper.ts b/src/gemini_wrapper.ts
new file mode 100644
index 0000000..bdfd440
--- /dev/null
+++ b/src/gemini_wrapper.ts
@@ -0,0 +1,64 @@
+import { GenerateContentRequest, VertexAI } from "@google-cloud/vertexai";
+// import util from "util";
+
+const projectId = "fleet-toolbox-404319";
+const location = "us-central1";
+const model = "gemini-1.0-pro";
+
+const vertexAI = new VertexAI({ project: projectId, location: location });
+const generativeModel = vertexAI.getGenerativeModel({
+  model: model,
+});
+
+function generateMessagesGemini(
+  input_text: string,
+  system: string,
+  examples: [string, string][]
+): GenerateContentRequest {
+  const examples_list = [];
+  for (const [user, response] of examples) {
+    examples_list.push(
+      { role: "user", parts: [{ text: user }] },
+      { role: "assistant", parts: [{ text: response }] }
+    );
+  }
+  return {
+    contents: [
+      { role: "system", parts: [{ text: system }] },
+      ...examples_list,
+      { role: "user", parts: [{ text: input_text }] },
+    ],
+  };
+}
+
+export async function queryGemini(
+  input_text: string,
+  system: string,
+  examples: [string, string][]
+): Promise<string> {
+  if (input_text.trim().length === 0) {
+    throw new TypeError("Input text can't be empty string");
+  }
+  const prompt = generateMessagesGemini(input_text, system, examples);
+  // TODO: add token counting
+
+  const resp = await generativeModel.countTokens(prompt);
+  console.log("count tokens response: ", resp);
+
+  // console.log(util.inspect(prompt, false, null, true));
+  // const prompt2: GenerateContentRequest = {
+  //   contents: [{ role: "user", parts: [{ text: "Tell me 3 dad jokes." }] }],
+  // };
+
+  try {
+    const response = await generativeModel.generateContent(prompt);
+    return response.response.candidates[0].content.parts[0].text || "";
+  } catch (error) {
+    if (error instanceof Error) {
+      console.error(`Error with Gemini API request, error: ${error}`);
+    } else {
+      console.error(`Error with Gemini API request: ${error}`);
+    }
+  }
+  return "";
+}
diff --git a/src/llm_runner.ts b/src/llm_runner.ts
index e56e10c..713cffb 100644
--- a/src/llm_runner.ts
+++ b/src/llm_runner.ts
@@ -1,43 +1,87 @@
-import { AsyncParser } from '@json2csv/node';
-import { program, Option, OptionValues } from 'commander';
+import { AsyncParser } from "@json2csv/node";
+import { program, Option, OptionValues } from "commander";
 
-import fs = require('node:fs/promises');
-import path = require('node:path');
+import fs = require("node:fs/promises");
+import path = require("node:path");
 
-import { INCENTIVES_FILE_BASE, OUTPUT_FILE_BASE, OUTPUT_SUBDIR, CSV_OPTS } from './constants.js';
-import { SYSTEM, EXAMPLE_1_RESPONSE, EXAMPLE_1_USER, EXAMPLE_2_RESPONSE, EXAMPLE_2_USER } from "./prompt.js"
+import {
+  INCENTIVES_FILE_BASE,
+  OUTPUT_FILE_BASE,
+  OUTPUT_SUBDIR,
+  CSV_OPTS,
+} from "./constants.js";
+import {
+  SYSTEM,
+  EXAMPLE_1_RESPONSE,
+  EXAMPLE_1_USER,
+  EXAMPLE_2_RESPONSE,
+  EXAMPLE_2_USER,
+} from "./prompt.js";
 
 import { queryPalm } from "./palm_wrapper.js";
 import { GptWrapper } from "./gpt_wrapper.js";
-import { Metadata } from "./metadata.js"
-import { generateHomeownerRenterField } from './post_processing/homeowner_renter.js';
-import { cleanupEnumFields } from './post_processing/cleanup_enum_fields.js';
-
+import { Metadata } from "./metadata.js";
+import { generateHomeownerRenterField } from "./post_processing/homeowner_renter.js";
+import { cleanupEnumFields } from "./post_processing/cleanup_enum_fields.js";
+import { queryGemini } from "./gemini_wrapper.js";
 
 program
-  .requiredOption("-f, --folders <folders...>", "Name of folder(s) under incentives_data/ where text data is located.")
-  .option("-r, --restrict <restrict_files...>", 'Will process only the files supplied. Useful for re-dos. Give the full path including INCENTIVES_FILE_BASE')
-  .option("-o, --output_file <file>", 'Name of output file. Saved in the out/ directory.', "output.csv")
-  .option("-w, --wait <duration_ms>", "How long to wait in ms between requests to avoid rate limiting")
-  .addOption(new Option("-m, --model_family <model_family>", 'Name of model family to use for queries').choices(['gpt4', 'gpt', 'palm']).default('palm'));
+  .requiredOption(
+    "-f, --folders <folders...>",
+    "Name of folder(s) under incentives_data/ where text data is located."
+  )
+  .option(
+    "-r, --restrict <restrict_files...>",
+    "Will process only the files supplied. Useful for re-dos. Give the full path including INCENTIVES_FILE_BASE"
+  )
+  .option(
+    "-o, --output_file <file>",
+    "Name of output file. Saved in the out/ directory.",
+    "output.csv"
+  )
+  .option(
+    "-w, --wait <duration_ms>",
+    "How long to wait in ms between requests to avoid rate limiting"
+  )
+  .addOption(
+    new Option(
+      "-m, --model_family <model_family>",
+      "Name of model family to use for queries"
+    )
+      .choices(["gpt4", "gpt", "palm", "gemini"])
+      .default("palm")
+  );
 
 program.parse();
 
-
-async function retrieveMetadata(folder: string, file: string): Promise<Metadata> {
-  const metadata_file = path.parse(file).name + "_metadata.json"
-  let contents: string = ""
+async function retrieveMetadata(
+  folder: string,
+  file: string
+): Promise<Metadata> {
+  const metadata_file = path.parse(file).name + "_metadata.json";
+  let contents: string = "";
   try {
-    contents = await fs.readFile(path.join(INCENTIVES_FILE_BASE, folder, metadata_file), { encoding: 'utf8' })
-    const metadata: Metadata = JSON.parse(contents)
-    return metadata
+    contents = await fs.readFile(
+      path.join(INCENTIVES_FILE_BASE, folder, metadata_file),
+      { encoding: "utf8" }
+    );
+    const metadata: Metadata = JSON.parse(contents);
+    return metadata;
   } catch (err) {
     if (err instanceof SyntaxError) {
-      console.log(`Error parsing metadata in ${metadata_file}: contents are ${contents}; error is ${err}`)
+      console.log(
+        `Error parsing metadata in ${metadata_file}: contents are ${contents}; error is ${err}`
+      );
     } else {
-      console.log(`No metadata file found: ${path.join(INCENTIVES_FILE_BASE, folder, file)}`);
+      console.log(
+        `No metadata file found: ${path.join(
+          INCENTIVES_FILE_BASE,
+          folder,
+          file
+        )}`
+      );
     }
-    return {}
+    return {};
   }
 }
 
@@ -45,8 +89,11 @@ function getParamsForLogging(opts: OptionValues) {
   return {
     model: opts.model_family,
     system: SYSTEM,
-    examples: [[EXAMPLE_1_USER, EXAMPLE_1_RESPONSE], [EXAMPLE_2_USER, EXAMPLE_2_RESPONSE]]
-  }
+    examples: [
+      [EXAMPLE_1_USER, EXAMPLE_1_RESPONSE],
+      [EXAMPLE_2_USER, EXAMPLE_2_RESPONSE],
+    ],
+  };
 }
 
 async function main() {
@@ -56,31 +103,44 @@ async function main() {
   const output: object[] = [];
   const metadata_fields: Set<string> = new Set<string>();
 
-  const runId = Date.now().toString()
+  const runId = Date.now().toString();
 
   await fs.mkdir(path.join(OUTPUT_FILE_BASE, runId));
-  await fs.writeFile(path.join(OUTPUT_FILE_BASE, runId, "parameters.json"), JSON.stringify(getParamsForLogging(opts)), {
-    encoding: "utf-8",
-    flag: "w"
-  });
+  await fs.writeFile(
+    path.join(OUTPUT_FILE_BASE, runId, "parameters.json"),
+    JSON.stringify(getParamsForLogging(opts)),
+    {
+      encoding: "utf-8",
+      flag: "w",
+    }
+  );
 
   await fs.mkdir(path.join(OUTPUT_FILE_BASE, runId, OUTPUT_SUBDIR));
   const droppedFiles: string[] = [];
   for (const folder of opts.folders) {
-    await fs.mkdir(path.join(OUTPUT_FILE_BASE, runId, OUTPUT_SUBDIR, folder)).catch(err => {
-      if (err.code !== 'EEXIST') {
-        console.log(err);
-      }
-    });
+    await fs
+      .mkdir(path.join(OUTPUT_FILE_BASE, runId, OUTPUT_SUBDIR, folder))
+      .catch((err) => {
+        if (err.code !== "EEXIST") {
+          console.log(err);
+        }
+      });
     const files = await fs.readdir(path.join(INCENTIVES_FILE_BASE, folder));
     for (const file of files) {
       if (!file.endsWith(".txt")) continue;
-      if (opts.restrict && !(opts.restrict.includes(path.join(INCENTIVES_FILE_BASE, folder, file)))) {
+      if (
+        opts.restrict &&
+        !opts.restrict.includes(path.join(INCENTIVES_FILE_BASE, folder, file))
+      ) {
         continue;
       }
-      const txt = (await fs.readFile(path.join(INCENTIVES_FILE_BASE, folder, file), { encoding: 'utf8' })).trim();
+      const txt = (
+        await fs.readFile(path.join(INCENTIVES_FILE_BASE, folder, file), {
+          encoding: "utf8",
+        })
+      ).trim();
       if (txt.length == 0) {
-        console.log(`Skipping ${path.join(folder, file)} because it is empty`)
+        console.log(`Skipping ${path.join(folder, file)} because it is empty`);
         continue;
       }
 
@@ -89,50 +149,87 @@ async function main() {
         metadata_fields.add(field);
       }
       if (metadata.tags !== undefined && metadata.tags.includes("index")) {
-        console.log(`Skipping ${path.join(folder, file)} because we detected an index tag`)
+        console.log(
+          `Skipping ${path.join(folder, file)} because we detected an index tag`
+        );
         continue;
       }
 
       if (opts.wait) {
-        await new Promise(f => setTimeout(f, +opts.wait))
+        await new Promise((f) => setTimeout(f, +opts.wait));
       }
 
-      console.log(`Querying ${opts.model_family} with ${path.join(INCENTIVES_FILE_BASE, folder, file)}`)
-      const gpt_wrapper = new GptWrapper(opts.model_family)
-      const queryFunc = opts.model_family == 'palm' ? queryPalm : gpt_wrapper.queryGpt.bind(gpt_wrapper)
-      const promise = queryFunc(txt, SYSTEM, [[EXAMPLE_1_USER, EXAMPLE_1_RESPONSE], [EXAMPLE_2_USER, EXAMPLE_2_RESPONSE]]).then(async msg => {
+      console.log(
+        `Querying ${opts.model_family} with ${path.join(
+          INCENTIVES_FILE_BASE,
+          folder,
+          file
+        )}`
+      );
+      const gpt_wrapper = new GptWrapper(opts.model_family);
+      let queryFunc;
+      switch (opts.model_family) {
+        case "palm": {
+          queryFunc = queryPalm;
+          break;
+        }
+        case "gemini": {
+          queryFunc = queryGemini;
+          break;
+        }
+        default: {
+          queryFunc = gpt_wrapper.queryGpt.bind(gpt_wrapper);
+          break;
+        }
+      }
+      const promise = queryFunc(txt, SYSTEM, [
+        [EXAMPLE_1_USER, EXAMPLE_1_RESPONSE],
+        [EXAMPLE_2_USER, EXAMPLE_2_RESPONSE],
+      ]).then(async (msg) => {
         if (msg == "") return;
-        console.log(`Got response from ${path.join(INCENTIVES_FILE_BASE, folder, file)}`)
+        console.log(
+          `Got response from ${path.join(INCENTIVES_FILE_BASE, folder, file)}`
+        );
         try {
           let records = JSON.parse(msg);
           if (!(Symbol.iterator in Object(records))) {
-            records = [records]
+            records = [records];
           }
 
           let incentive_order_key = 0;
-          const file_records: object[] = []
+          const file_records: object[] = [];
           let combined: object = {};
           for (const record of records) {
-            cleanupEnumFields(record)
-            generateHomeownerRenterField(record)
+            cleanupEnumFields(record);
+            generateHomeownerRenterField(record);
 
-            if (!('folder' in metadata)) {
-              metadata['folder'] = folder;
+            if (!("folder" in metadata)) {
+              metadata["folder"] = folder;
             }
-            metadata['file'] = file; // For debugging.
-            metadata['order'] = incentive_order_key;
+            metadata["file"] = file; // For debugging.
+            metadata["order"] = incentive_order_key;
             combined = { ...record, ...metadata };
             output.push(combined);
-            file_records.push(combined)
+            file_records.push(combined);
             incentive_order_key += 1;
           }
-          await fs.writeFile(path.join(OUTPUT_FILE_BASE, runId, OUTPUT_SUBDIR, folder, file.replace(".txt", "_output.json")), JSON.stringify(file_records, null, 2), {
-            encoding: "utf-8",
-            flag: "w"
-          })
+          await fs.writeFile(
+            path.join(
+              OUTPUT_FILE_BASE,
+              runId,
+              OUTPUT_SUBDIR,
+              folder,
+              file.replace(".txt", "_output.json")
+            ),
+            JSON.stringify(file_records, null, 2),
+            {
+              encoding: "utf-8",
+              flag: "w",
+            }
+          );
         } catch (error) {
           console.error(`Error parsing json: ${error}, ${msg}`);
-          droppedFiles.push(path.join(INCENTIVES_FILE_BASE, folder, file))
+          droppedFiles.push(path.join(INCENTIVES_FILE_BASE, folder, file));
         }
       });
       promises.push(promise);
@@ -147,13 +244,23 @@ async function main() {
 
     const parser = new AsyncParser(CSV_OPTS);
     const csv = await parser.parse(output).promise();
-    await fs.writeFile(path.join(OUTPUT_FILE_BASE, runId, opts.output_file), csv);
+    await fs.writeFile(
+      path.join(OUTPUT_FILE_BASE, runId, opts.output_file),
+      csv
+    );
     if (droppedFiles.length > 0) {
-      await fs.writeFile(path.join(OUTPUT_FILE_BASE, runId, "dropped_files.json"), JSON.stringify(droppedFiles));
+      await fs.writeFile(
+        path.join(OUTPUT_FILE_BASE, runId, "dropped_files.json"),
+        JSON.stringify(droppedFiles)
+      );
     }
-    console.log(`Find your results with run ID ${runId} at ${path.join(OUTPUT_FILE_BASE, runId)}`)
-
-  })
+    console.log(
+      `Find your results with run ID ${runId} at ${path.join(
+        OUTPUT_FILE_BASE,
+        runId
+      )}`
+    );
+  });
 }
 
-main()
\ No newline at end of file
+main();
diff --git a/src/test_request_gemini.ts b/src/test_request_gemini.ts
new file mode 100644
index 0000000..c6aac2a
--- /dev/null
+++ b/src/test_request_gemini.ts
@@ -0,0 +1,40 @@
+import { VertexAI } from "@google-cloud/vertexai";
+
+/**
+ * TODO(developer): Update these variables before running the sample.
+ */
+async function createNonStreamingMultipartContent(
+  projectId = "fleet-toolbox-404319",
+  location = "us-central1",
+  model = "gemini-1.0-pro"
+) {
+  // Initialize Vertex with your Cloud project and location
+  const vertexAI = new VertexAI({ project: projectId, location: location });
+
+  // Instantiate the model
+  const generativeVisionModel = vertexAI.getGenerativeModel({
+    model: model,
+  });
+
+  const textPart = {
+    text: "You are a helpful assistant. Give me three dad jokes.",
+  };
+
+  const request = {
+    contents: [{ role: "user", parts: [textPart] }],
+  };
+
+  console.log("Prompt Text:");
+  console.log(request.contents[0].parts[0].text);
+
+  console.log("Non-Streaming Response Text:");
+  const response = await generativeVisionModel.generateContent(request);
+
+  // Select the text from the response
+  const fullTextResponse =
+    response.response.candidates[0].content.parts[0].text;
+
+  console.log(fullTextResponse);
+}
+
+createNonStreamingMultipartContent();