feat(labs): PDF upload, text extraction, LLM draft pipeline

server/src/app.ts

@@ -3,9 +3,10 @@ import { getCookie } from "hono/cookie";
 import { serveStatic } from "hono/bun";
 import type { Db } from "./db";
 import { authRoutes, isAuthenticated } from "./routes/auth";
+import { labsRoutes } from "./routes/labs";
 import { settingsRoutes } from "./routes/settings";
 
-export type Deps = { db: Db; key: Buffer };
+export type Deps = { db: Db; key: Buffer; dataDir: string; llmFetch?: typeof fetch };
 const PUBLIC = new Set(["/api/health", "/api/me", "/api/setup", "/api/login"]);
 
 export function createApp(deps: Deps) {
@@ -20,6 +21,7 @@ export function createApp(deps: Deps) {
   });
   app.route("/api", authRoutes({ db: deps.db }));
   app.route("/api", settingsRoutes(deps));
+  app.route("/api", labsRoutes(deps));
   // Later route groups (connectors, chat) mount here.
   app.all("/api/*", (c) => c.json({ error: "not found" }, 404));
   app.use("/*", serveStatic({ root: "./web/dist" }));

server/src/index.ts

@@ -3,6 +3,6 @@ import { loadOrCreateKey } from "./lib/crypto";
 import { createApp } from "./app";
 
 const dataDir = process.env.DATA_DIR ?? "./data";
-const app = createApp({ db: openDb(dataDir), key: loadOrCreateKey(dataDir) });
+const app = createApp({ db: openDb(dataDir), key: loadOrCreateKey(dataDir), dataDir });
 
 export default { port: Number(process.env.PORT ?? 3000), fetch: app.fetch };

server/src/lib/extract.ts

@@ -0,0 +1,15 @@
+import { ExtractedDraft } from "@helios/shared";
+import { chatJSON, type LlmDeps } from "./llm";
+
+const SYSTEM = `You extract structured blood-test results from the raw text of a lab report.
+Return ONLY a JSON object: {"collectedDate": "YYYY-MM-DD" or null, "labName": string or null, "markers": [{"panel": string or null, "name": string, "value": string, "unit": string or null, "referenceRange": string or null, "flagged": boolean}]}.
+Rules: copy names, values, units, and reference ranges EXACTLY as printed — do not convert units or round values. "value" is always a string (keep comparators like "<0.3"). Set "flagged" true only when the report marks the result abnormal (H, L, *, bold, out-of-range annotation). Use the specimen collection date, not the report date. Skip commentary, footers, and reference-only rows with no result.`;
+
+const MAX_CHARS = 40_000;
+
+export async function extractFromText(deps: LlmDeps, text: string): Promise<ExtractedDraft> {
+  const out = await chatJSON(deps, { system: SYSTEM, user: text.slice(0, MAX_CHARS) });
+  const parsed = ExtractedDraft.safeParse(out);
+  if (!parsed.success) throw new Error("llm_error: draft failed schema validation");
+  return parsed.data;
+}

server/src/lib/pdf.ts

@@ -0,0 +1,7 @@
+import { extractText, getDocumentProxy } from "unpdf";
+
+export async function pdfToText(data: Uint8Array): Promise<string> {
+  const doc = await getDocumentProxy(data);
+  const { text } = await extractText(doc, { mergePages: true });
+  return text;
+}

server/src/routes/labs.ts

@@ -0,0 +1,68 @@
+import { desc, eq } from "drizzle-orm";
+import { Hono } from "hono";
+import { randomUUID } from "node:crypto";
+import { mkdirSync } from "node:fs";
+import { join } from "node:path";
+import type { Db } from "../db";
+import { labDrafts } from "../db/schema";
+import { extractFromText } from "../lib/extract";
+import { pdfToText } from "../lib/pdf";
+
+export type LabsDeps = { db: Db; key: Buffer; dataDir: string; llmFetch?: typeof fetch };
+
+const MAX_UPLOAD = 15 * 1024 * 1024;
+
+export function labsRoutes(deps: LabsDeps) {
+  const app = new Hono();
+
+  app.post("/labs/upload", async (c) => {
+    const body = await c.req.parseBody();
+    const file = body.file;
+    if (!(file instanceof File)) return c.json({ error: "file field required" }, 400);
+    if (!file.name.toLowerCase().endsWith(".pdf") && file.type !== "application/pdf") {
+      return c.json({ error: "not_a_pdf" }, 400);
+    }
+    if (file.size > MAX_UPLOAD) return c.json({ error: "too_large" }, 400);
+
+    const id = randomUUID();
+    const uploadsDir = join(deps.dataDir, "uploads");
+    mkdirSync(uploadsDir, { recursive: true });
+    const filePath = join(uploadsDir, `${id}.pdf`);
+    const bytes = new Uint8Array(await file.arrayBuffer());
+    await Bun.write(filePath, bytes);
+
+    // Extraction failures land on the draft row so the user sees them in the
+    // review UI instead of the upload 500ing.
+    let extracted: string | null = null;
+    let error: string | null = null;
+    try {
+      const text = await pdfToText(bytes);
+      if (text.trim().length < 20) throw new Error("pdf_error: no text layer (scanned PDFs are not supported yet)");
+      const draft = await extractFromText({ db: deps.db, key: deps.key, fetchImpl: deps.llmFetch }, text);
+      extracted = JSON.stringify(draft);
+    } catch (e) {
+      error = e instanceof Error ? e.message : "extraction failed";
+    }
+
+    await deps.db.insert(labDrafts).values({
+      id, filename: file.name, filePath, status: "pending", extracted, error, createdAt: Date.now(),
+    });
+    return c.json({ id }, 201);
+  });
+
+  app.get("/labs/drafts", async (c) => {
+    const rows = await deps.db.select().from(labDrafts).orderBy(desc(labDrafts.createdAt));
+    return c.json({
+      drafts: rows.map((r) => ({
+        id: r.id,
+        filename: r.filename,
+        status: r.status,
+        error: r.error,
+        markerCount: r.extracted ? (JSON.parse(r.extracted).markers?.length ?? 0) : 0,
+        createdAt: r.createdAt,
+      })),
+    });
+  });
+
+  return app;
+}