Mohammad Nasim — Portfolio & CV Chatbot

Personal portfolio site at mohammadnasim.com with an embedded AI chatbot that answers visitor questions about Mohammad's background, grounded strictly in his CV via retrieval-augmented generation (RAG).

The repo holds both the static frontend (HTML/CSS/JS deployed to Hostinger) and the Python backend (FastAPI + LangChain on Google Cloud Run).

Architecture at a glance

Live URLs

Site: https://mohammadnasim.com
Chatbot backend: https://cv-chatbot-679776566774.us-central1.run.app
Backend /health: returns {"status":"ok","chunks_indexed":N}

Tech stack

Layer	Stack
Frontend	Vanilla HTML5, CSS3, ES6+ JavaScript (no framework)
Backend API	Python 3.11, FastAPI, Uvicorn, Pydantic
RAG	LangChain (LCEL), OpenAI embeddings, FAISS vector store, PyPDF
LLM	OpenAI `gpt-5.4-mini` (temp 0.2)
Container	Docker (python:3.11-slim base, libgomp1 for FAISS)
Hosting	Google Cloud Run (backend), Hostinger FTP (frontend)
CI/CD	GitHub Actions (frontend FTP deploy on push to main)
Secrets	GCP Secret Manager (OpenAI key bound at deploy time)

Repository layout

.
├── index.html                       # Single-page portfolio site
├── styles.css                       # Site styles (dark/light theme, animations)
├── script.js                        # Chat widget + site behaviors
├── image/                           # Hero photo and static assets
│
├── chatbot_backend.py               # FastAPI app + RAG pipeline
├── requirements.txt                 # Python deps
├── Dockerfile                       # Container build for Cloud Run
├── .dockerignore                    # Excludes frontend assets from image
│
├── cv/
│   └── AgenticAI Architect Mohammad Nasim V5.0_Detailed.pdf
│
├── faiss_index/                     # Pre-built vector store (baked into image)
│   ├── index.faiss
│   └── index.pkl
│
├── .github/workflows/
│   └── deploy-frontend.yml          # Hostinger FTP deploy
│
└── .env                             # Local-only: OPENAI_API_KEY (gitignored ideally)

How the chatbot works

Indexing pipeline (runs once per container build)

Load — PyPDFLoader reads the CV PDF and returns one Document per page.
Annotate — a custom annotate_employer() walks the concatenated text top-to-bottom, detecting employer headers (e.g. "State of Rhode Island Department of Transportation", "Atkins UK", "PAEW") and prepending [Employer: <name>] to every line in that section. This guarantees each chunk carries explicit attribution regardless of where the splitter cuts.
Split — RecursiveCharacterTextSplitter with chunk_size=2000, overlap=400, separators ["\n\n", "\n", ".", " ", ""].
Embed — OpenAIEmbeddings produces vectors for each chunk.
Persist — FAISS.save_local() writes index.faiss and index.pkl to disk. These files are baked into the Docker image so cold starts skip the embedding step.

Query pipeline (per request)

POST /chat with {"message": "...", "session_id": "..."}.
Retriever pulls top-k=8 chunks from FAISS by cosine similarity to the question.
The chunks are joined into a CV Context: block and passed to the LLM along with the chat history for that session.
The LLM is instructed to answer strictly from context, refuse fabrication, attribute projects only to employers the context explicitly ties them to, and keep responses under 200 words.
Response is returned and appended to the session history (capped at 6 turn pairs).

Why the employer annotation matters

Without it, an 800-char chunk like "Projects: 1. AI-Powered Bridge Monitoring..." had no client name attached, so when retrieved for a Rhode Island question the LLM would misattribute it to Rhode Island. Tagging every line with [Employer: State of Michigan Department of Transportation] makes the attribution travel with the text into every chunk it lands in.

Local development

Prerequisites

Python 3.11+ (3.13 works; 3.11 is what Docker uses)
A working OpenAI API key with sufficient quota
(Optional) gcloud CLI if you plan to deploy

Setup

# Clone (already done if reading this from the repo)
git clone https://github.com/MohammadNasim/Portfolio2.git
cd Portfolio2

# Create a venv and install deps
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

# Add your OpenAI key
"OPENAI_API_KEY=sk-proj-..." | Set-Content .env

# Run
python chatbot_backend.py
# → http://localhost:8000/health
# → http://localhost:8000/chat (POST)

On first run the FAISS index is loaded from faiss_index/ on disk (built from V5 PDF). Delete that directory to force a rebuild on next startup.

Frontend

The frontend is a single index.html — open it directly in a browser or serve it locally:

python -m http.server 5500
# → http://localhost:5500/index.html

To point the chat widget at a local backend, edit script.js:

const CHATBOT_API_URL = "http://localhost:8000";

(remember to revert before pushing — production points at the Cloud Run URL).

Production deployment

One-time GCP setup (already done for the production project)

$PROJECT = "prjadkvertexairagservice"
$REGION  = "us-central1"

# Enable APIs
gcloud services enable run.googleapis.com artifactregistry.googleapis.com cloudbuild.googleapis.com secretmanager.googleapis.com --project=$PROJECT

# Artifact Registry repo
gcloud artifacts repositories create chatbot --repository-format=docker --location=$REGION --project=$PROJECT

# Cloud Build runner SA (newer GCP projects lack the legacy default)
gcloud iam service-accounts create cloudbuild-runner --project=$PROJECT
gcloud projects add-iam-policy-binding $PROJECT --member="serviceAccount:cloudbuild-runner@$PROJECT.iam.gserviceaccount.com" --role="roles/cloudbuild.builds.builder"
gcloud projects add-iam-policy-binding $PROJECT --member="serviceAccount:cloudbuild-runner@$PROJECT.iam.gserviceaccount.com" --role="roles/artifactregistry.writer"
gcloud projects add-iam-policy-binding $PROJECT --member="serviceAccount:cloudbuild-runner@$PROJECT.iam.gserviceaccount.com" --role="roles/logging.logWriter"

# Cloud Run runtime SA + secret access
gcloud iam service-accounts create chatbot-runtime --project=$PROJECT
gcloud secrets create OPENAI_API_KEY --replication-policy=automatic --project=$PROJECT
"sk-proj-..." | gcloud secrets versions add OPENAI_API_KEY --data-file=- --project=$PROJECT
gcloud secrets add-iam-policy-binding OPENAI_API_KEY --member="serviceAccount:chatbot-runtime@$PROJECT.iam.gserviceaccount.com" --role="roles/secretmanager.secretAccessor" --project=$PROJECT

Build and deploy (every release)

$IMAGE = "us-central1-docker.pkg.dev/prjadkvertexairagservice/chatbot/cv-chatbot:latest"

# 1. Build image via Cloud Build
gcloud builds submit `
  --tag $IMAGE `
  --project=prjadkvertexairagservice `
  --region=us-central1 `
  --service-account=projects/prjadkvertexairagservice/serviceAccounts/cloudbuild-runner@prjadkvertexairagservice.iam.gserviceaccount.com `
  --default-buckets-behavior=REGIONAL_USER_OWNED_BUCKET

# 2. Roll new Cloud Run revision (env vars + secret bindings persist from prior deploy)
gcloud run deploy cv-chatbot --image $IMAGE --region us-central1 --project=prjadkvertexairagservice

Initial deploy used these flags (only needed once; subsequent deploys inherit them):

gcloud run deploy cv-chatbot `
  --image $IMAGE `
  --region us-central1 `
  --platform managed `
  --allow-unauthenticated `
  --service-account chatbot-runtime@prjadkvertexairagservice.iam.gserviceaccount.com `
  --memory 1Gi --cpu 1 `
  --min-instances 0 --max-instances 3 `
  --timeout 300 `
  "--set-env-vars=^##^ALLOWED_ORIGINS=https://mohammadnasim.com,https://www.mohammadnasim.com" `
  --set-secrets "OPENAI_API_KEY=OPENAI_API_KEY:latest"

The ^##^ delimiter is required because ALLOWED_ORIGINS contains commas — gcloud uses commas as the env-var separator by default.

Frontend deploy (automatic)

Pushing to main triggers .github/workflows/deploy-frontend.yml, which FTPs index.html, styles.css, script.js, and image/** to Hostinger's public_html/. Backend-only changes (chatbot_backend.py, Dockerfile, faiss_index/) do not trigger the workflow — only index.html / script.js / styles.css / image/** paths do.

Common operations

Rebuild FAISS index from updated CV

# Replace the PDF in cv/ first, then:
Remove-Item -Recurse -Force faiss_index
$env:OPENAI_API_KEY = "sk-proj-..."
python -c "import chatbot_backend"   # rebuilds + saves to disk

# Commit, then redeploy (see "Build and deploy" above)
git add faiss_index chatbot_backend.py cv/
git commit -m "Update CV and rebuild FAISS"

Rotate the OpenAI key on Cloud Run

# Add new version to Secret Manager
printf '%s' 'sk-proj-NEW_KEY' | gcloud secrets versions add OPENAI_API_KEY --data-file=- --project=prjadkvertexairagservice

# Pin the service to the new version (this also forces a new revision)
gcloud run services update cv-chatbot --region us-central1 --project=prjadkvertexairagservice --update-secrets "OPENAI_API_KEY=OPENAI_API_KEY:<N>"

Inspect current deployment state

# Active revision and config
gcloud run services describe cv-chatbot --region us-central1 --project=prjadkvertexairagservice

# Image history in Artifact Registry
gcloud artifacts docker images list us-central1-docker.pkg.dev/prjadkvertexairagservice/chatbot/cv-chatbot --include-tags --sort-by=~UPDATE_TIME

# Recent Cloud Build runs
gcloud builds list --project=prjadkvertexairagservice --limit=10

# Live smoke test
$body = @{ message = "Who is Mohammad?"; session_id = "test" } | ConvertTo-Json
Invoke-RestMethod -Uri https://cv-chatbot-679776566774.us-central1.run.app/chat -Method Post -ContentType "application/json" -Body $body

Adjust system prompt or chunking

Edit chatbot_backend.py:

SYSTEM_MSG (around line 90) — instructions to the LLM
EMPLOYER_ANCHORS and annotate_employer() — attribution preprocessing
RecursiveCharacterTextSplitter(...) args — chunk size and overlap
vectorstore.as_retriever(search_kwargs={"k": 8}) — number of chunks retrieved

Then rebuild FAISS (delete faiss_index/ and re-run) and redeploy.

API reference

`GET /health`

{"status": "ok", "chunks_indexed": 23}

`POST /chat`

Request

{ "message": "What is Mohammad's current role?", "session_id": "user-123" }

Response (200)

{
  "answer": "Mohammad is currently...",
  "session_id": "user-123"
}

Error responses

400 — empty message
500 — upstream LLM/embedding error (passes through OpenAI error detail)

Sessions are in-memory only — they reset when a Cloud Run instance dies or scales to zero. Each session retains the last 6 human/AI turn pairs.

`DELETE /session/{session_id}`

Clears history for the given session.

Configuration

Environment variables (Cloud Run)

Variable	Source	Purpose
`OPENAI_API_KEY`	Secret Manager	OpenAI API authentication
`ALLOWED_ORIGINS`	`--set-env-vars`	Comma-separated CORS allowlist
`PORT`	Cloud Run	Injected automatically; uvicorn binds here

Local `.env`

OPENAI_API_KEY=sk-proj-...
ALLOWED_ORIGINS=http://localhost:5500

.env is loaded by python-dotenv only — Cloud Run ignores it and uses the Secret Manager binding instead.

Troubleshooting

Symptom	Likely cause	Fix
Chat widget shows "Could not reach the AI backend"	CORS blocked or Cloud Run cold-starting	Verify `ALLOWED_ORIGINS` includes the calling origin; first request after idle takes 5–20s
`/chat` returns `insufficient_quota` 500	OpenAI billing or rate limit	Top up OpenAI billing or check usage caps
Chatbot says "Architect" instead of "Developer"	Old image still deployed	Run `gcloud builds submit` + `gcloud run deploy` again
`gcloud builds submit` fails with `NOT_FOUND`	Cloud Build default SA missing	Use `--service-account=...cloudbuild-runner...` + `--default-buckets-behavior=REGIONAL_USER_OWNED_BUCKET`
Cross-employer hallucination in answers	Chunk boundaries split client from project list	Verify `annotate_employer()` is running before splitting; check that anchor strings match the PDF text
Frontend not updating after push	Workflow didn't trigger	Confirm push touched `index.html`/`script.js`/`styles.css`/`image/**` (only those paths fire the deploy)

Project resource map

For day-to-day operations, here's where each piece of state lives:

Resource	Location
GCP project	`prjadkvertexairagservice` (project number 679776566774)
Cloud Run service	`cv-chatbot` in `us-central1`
Docker image	`us-central1-docker.pkg.dev/prjadkvertexairagservice/chatbot/cv-chatbot:latest`
Build service account	`cloudbuild-runner@prjadkvertexairagservice.iam.gserviceaccount.com`
Runtime service account	`chatbot-runtime@prjadkvertexairagservice.iam.gserviceaccount.com`
OpenAI key secret	`OPENAI_API_KEY` in Secret Manager
Billing account	`0102B6-A64631-55DF6A` ("Nasim Portfolio")
Frontend host	Hostinger, deployed via GitHub Actions FTP
Source repo	https://github.com/MohammadNasim/Portfolio2

License & contact

This is a personal portfolio — code is provided as-is for reference. The CV PDF and likeness are property of Mohammad Nasim.

For questions: mnasimsiddiqui@gmail.com · LinkedIn

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Mohammad Nasim — Portfolio & CV Chatbot

Architecture at a glance

Tech stack

Repository layout

How the chatbot works

Indexing pipeline (runs once per container build)

Query pipeline (per request)

Why the employer annotation matters

Local development

Prerequisites

Setup

Frontend

Production deployment

One-time GCP setup (already done for the production project)

Build and deploy (every release)

Frontend deploy (automatic)

Common operations

Rebuild FAISS index from updated CV

Rotate the OpenAI key on Cloud Run

Inspect current deployment state

Adjust system prompt or chunking

API reference

`GET /health`

`POST /chat`

`DELETE /session/{session_id}`

Configuration

Environment variables (Cloud Run)

Local `.env`

Troubleshooting

Project resource map

License & contact

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 29 Commits
.github/workflows		.github/workflows
__pycache__		__pycache__
cv		cv
faiss_index		faiss_index
image		image
.dockerignore		.dockerignore
.env		.env
.gitattributes		.gitattributes
Dockerfile		Dockerfile
PortfolioArchitecture.png		PortfolioArchitecture.png
README.md		README.md
chatbot_backend.py		chatbot_backend.py
index.html		index.html
requirements.txt		requirements.txt
script.js		script.js
styles.css		styles.css

Folders and files

Latest commit

History

Repository files navigation

Mohammad Nasim — Portfolio & CV Chatbot

Architecture at a glance

Tech stack

Repository layout

How the chatbot works

Indexing pipeline (runs once per container build)

Query pipeline (per request)

Why the employer annotation matters

Local development

Prerequisites

Setup

Frontend

Production deployment

One-time GCP setup (already done for the production project)

Build and deploy (every release)

Frontend deploy (automatic)

Common operations

Rebuild FAISS index from updated CV

Rotate the OpenAI key on Cloud Run

Inspect current deployment state

Adjust system prompt or chunking

API reference

GET /health

POST /chat

DELETE /session/{session_id}

Configuration

Environment variables (Cloud Run)

Local .env

Troubleshooting

Project resource map

License & contact

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

`GET /health`

`POST /chat`

`DELETE /session/{session_id}`

Local `.env`