Indorex AI — Datathon Polyfinances

AI assistant that ingests new regulatory/fiscal texts, analyzes their impact on an equity portfolio (S&P-500 reference), and returns portfolio-aware recommendations. Stack: React (Vite) + API Gateway (REST) + AWS Lambda (Python) + Bedrock (Claude Sonnet), optional Amazon Kendra for retrieval.

---

1) Features

• 🔼 Upload regulatory docs (CSV/JSON/PDF UI; file metadata currently used, full ingest next)
• 🧠 Prompt includes user context + holdings (from src/data/portfolio.json)
• 🔎 (Planned) Retrieve supporting evidence via Amazon Kendra
• 🌐 Fully CORS-enabled API (browser friendly)
• ☁️ Serverless backend (Lambda) with Bedrock model call

---

2) High-level Architecture

React UI (Vite) ── POST /prod/ ──► API Gateway (REST)
                                   │
                                   └─► Lambda (Python)
                                         │
                                         └─► Bedrock Runtime (Claude 3.5 Sonnet)
                                              (Optional) Kendra for retrieval
S3 ──(Kendra data source)──► Kendra Index

---

3) Repos & Key Paths

UI/
  src/pages/ingestion/index.tsx   # Ingestion page (frontend)
  src/data/portfolio.json         # Sample holdings used in prompts
lambda/
  lambda_function.py              # Backend handler (Bedrock + CORS)

---

4) AWS Resources (current)

• API Gateway (REST): 1pixxoj603 → Stage: prod

  • Methods on root /: OPTIONS (Mock), POST (Lambda)
  • Important: Frontend must call https://1pixxoj603.execute-api.us-east-1.amazonaws.com/prod/ (note trailing /)

• Lambda: indorex-ai-backend (role: indorex-ai-backend-role-v9qcn1bg)

• Bedrock Model: anthropic.claude-3-5-sonnet-20240620-v1:0 (us-east-1)

• Kendra Index: datathon-polyfinances-index-b9b3e5fc Data source (S3): datathon-polyfinances-s3-source

• S3 examples: 2025-09-26_stocks-performance.csv, 2025-08-15_composition_sp500.csv

---

5) Prerequisites

• Node 18+

• AWS account with access to:

  • Bedrock in us-east-1 (Model Access enabled for Claude 3.5 Sonnet)
  • Lambda, API Gateway, S3, Kendra (optional)

---

6) Local Development

UI

cd UI
npm i
npm run dev
# Open http://localhost:5173/dashboard

Keep the trailing slash to avoid MissingAuthenticationToken.

Lambda (Python)

• Package lambda/lambda_function.py as a ZIP and upload, or use your pipeline.
• Runtime: Python 3.10+ (compatible with boto3/botocore in AWS runtime)

---

7) Bedrock Permissions (must do)

A) Enable model access in console

Bedrock → Model access → enable Claude 3.5 Sonnet in us-east-1.

B) Lambda execution role policy

Attach an inline policy to indorex-ai-backend-role-v9qcn1bg:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "AllowClaudeInvoke",
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:InvokeModelWithResponseStream"
      ],
      "Resource": "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-3-5-sonnet-20240620-v1:0"
    },
    {
      "Sid": "AllowMarketplaceSubsView",
      "Effect": "Allow",
      "Action": "aws-marketplace:ViewSubscriptions",
      "Resource": "*"
    }
  ]
}

If you just added this, give IAM/STS a few minutes, then redeploy the Lambda once.

---

8) API Contract

Request

POST https://1pixxoj603.execute-api.us-east-1.amazonaws.com/prod/

{
  "message": "You are a financial analysis assistant...\nInput JSON: { context, portfolio, files }"
}

Frontend composes message including:

• context (textarea)
• portfolio (from src/data/portfolio.json)
• files (array of file metadata; full upload pipeline pending)

Response (success)

{
  "answer": "Model-generated analysis..."
}

CORS (preflight)

OPTIONS /prod/ returns:

Access-Control-Allow-Origin: *
Access-Control-Allow-Methods: GET,POST,OPTIONS
Access-Control-Allow-Headers: Content-Type,X-Amz-Date,Authorization,X-Api-Key,X-Amz-Security-Token

---

9) Quick Tests

Preflight:

curl -i -X OPTIONS https://1pixxoj603.execute-api.us-east-1.amazonaws.com/prod/ \
  -H "Origin: http://localhost:5173" \
  -H "Access-Control-Request-Method: POST"

Invoke:

curl -i -X POST https://1pixxoj603.execute-api.us-east-1.amazonaws.com/prod/ \
  -H "Content-Type: application/json" \
  -d '{"message":"AI portfolio analysis"}'

---

10) Kendra (optional, next)

• Index: datathon-polyfinances-index-b9b3e5fc (Active)
• Data Source: datathon-polyfinances-s3-source (S3 sync working)
• Next: Add a Lambda step to query Kendra and inject top passages into the Bedrock prompt.

---

11) Troubleshooting

Failed to fetch (browser) / CORS

• Verify preflight returns 200 with the headers above.
• Ensure API URL has trailing slash (/prod/).

MissingAuthenticationToken

• Usually hitting /prod (no trailing slash). Use /prod/.

AccessDeniedException bedrock:InvokeModel or aws-marketplace:ViewSubscriptions

• Attach the IAM policy above and ensure Bedrock Model Access is enabled for the account.
• Wait ~5–15 minutes for STS cache; redeploy Lambda.

NoneType has no attribute get in Lambda

• Ensure you parse event["body"] safely before using.

---

12) Security & Cost Notes

• Lock down IAM to least privilege once stabilized (narrow resources where possible).
• Bedrock/Kendra incur usage costs—monitor in Cost Explorer.

---

13) Roadmap

• ✅ CORS stable, UI ↔ API ↔ Lambda OK
• 🚧 Bedrock invocation: finish permissions + model access, return real AI analysis
• 🔜 Wire Kendra retrieval (RAG)
• 🔜 Real file uploads to S3 + parsing pipeline
• 🔜 Better output formatting (risk metrics, sector exposure, recs)

---

14) Maintainers

• Indorex AI / Datathon Polyfinances team
• AWS stack: us-east-1
• Primary Lambda: indorex-ai-backend
• API: 1pixxoj603 (stage prod)