AI Integration and Custom Development & Business Process Automation

We're a growing B2B services company looking for an experienced team (or senior full-stack developer) who can do two things at once: • Hardcore custom development — build a lightweight internal web app, custom APIs, and a database layer that ties our tools together where no-code can't reach. • Business process automation — design and deploy AI-powered workflows across sales, marketing, and customer support using Make/n8n + LLM APIs. Right now our team is losing 15–20 hours a week to manual work — copying leads between tools, writing repetitive replies, updating our CRM, qualifying inbound inquiries, and pulling reports. Off-the-shelf tools get us 70% of the way there; the last 30% needs actual code. Current stack: HubSpot (CRM), Gmail/Google Workspace, Slack, Calendly, Google Sheets, Notion. Open to adding Make, n8n, OpenAI/Claude API, Bland.AI, Supabase, and a custom Node.js/Python backend where it makes sense. Part A — Hardcore Development Scope This is the part that separates real engineers from Zapier hobbyists. We need production-grade code, not glue together demos. 1. Custom Middleware / API Layer • Build a Node.js (or Python/FastAPI) backend hosted on AWS, GCP, or Vercel • Expose secure REST endpoints for our internal tools and automation platform to call • Handle authentication (JWT / API keys), rate limiting, logging, and error retries • Webhook receivers for HubSpot, Calendly, Stripe, and inbound email • Written with clean architecture — controllers, services, repositories — not spaghetti scripts 2. Custom Database Layer • PostgreSQL (Supabase or AWS RDS) for storing enriched lead data, conversation history, LLM outputs, and audit logs • Proper schema design with migrations (Prisma / Drizzle / Alembic) • Row-level security where relevant 3. Internal Admin Dashboard (Web App) • Next.js or React admin panel where our sales/ops team can: ◦ Review AI-drafted emails before send ◦ Approve/reject qualified leads ◦ View automation run history and error logs ◦ Manually re-run failed jobs • Auth via Google Workspace SSO • Responsive, clean UI (Tailwind + shadcn/ui preferred) 4. LLM Orchestration Layer (Custom Code) • A Python or TypeScript service that wraps OpenAI/Claude calls with: ◦ Prompt templates + versioning ◦ Structured output validation (Zod / Pydantic) ◦ Retry logic on malformed responses ◦ Token usage logging and cost tracking per workflow • Lightweight RAG using pgvector for pulling company-specific context into prompts 5. DevOps & Deployment • Git repo (GitHub) with clean commit history and README • CI/CD via GitHub Actions • Staging + production environments • Environment variable and secrets management (Doppler / AWS Secrets Manager) • Monitoring: error tracking (Sentry) + uptime alerts Part B — Business Process Automation Scope Once the backend and admin panel exist, we wire the workflows on top. 1. AI-Powered Lead Qualification • Capture inbound leads from website form, LinkedIn, and cold email replies • Auto-enrich via Apollo or Clearbit API • LLM scores and categorizes leads (Hot / Warm / Cold) against our ICP — logic lives in the custom LLM service, not inside Make • Push qualified leads into HubSpot with the correct pipeline stage and owner 2. AI Email Response Assistant • Draft first-touch replies using our tone and past conversation context (pulled from PostgreSQL history) • Route to the admin dashboard for one-click human review before sending • Log every interaction back to the HubSpot contact timeline 3. Sales Ops Automation • Auto-create HubSpot deals when a meeting is booked via Calendly • Auto-generate meeting prep notes (company summary, recent news, LinkedIn highlights) 15 minutes before each call, delivered to Slack • Post-meeting: transcribe (Fathom/Fireflies), summarize action items, update the CRM 4. AI Voice Follow-Up (Bland.AI or Vapi) • Automated outbound voice call to re-engage cold leads that haven't replied in 14 days • Call script generated dynamically from lead's history and enrichment data • Transcript + summary logged to HubSpot and admin dashboard 5. Reporting & Internal Dashboards • Weekly automated pipeline pulling HubSpot + Google Ads + GA4 into Looker Studio • AI-generated executive summary (3–4 bullets) sent to Slack every Monday 9 AM • Same data also visible inside our internal admin dashboard Deliverables • Deployed backend + database + admin dashboard (source code + hosted URLs) • All five automation workflows live and running • GitHub repo with README, setup instructions, and architecture diagram • Loom walkthroughs + written SOPs for each workflow • 2-hour handover + training session (recorded) • 21 days of post-launch support for bug fixes and minor tweaks • Phase 2 roadmap doc — what we should build next Budget & Timeline Budget: $3,000 USD (fixed price) Timeline: 5–6 weeks from kickoff Milestones: • M1 — Discovery, architecture doc, DB schema, repo setup ($400) • M2 — Backend API + LLM orchestration service + admin dashboard skeleton ($800) • M3 — Lead qualification + Email assistant workflows wired end-to-end ($700) • M4 — Sales ops + Voice follow-up + Reporting workflows ($700) • M5 — Documentation, handover, 21-day support ($400) Required Skills & Expertise Development (must-have) • Node.js / TypeScript or Python (FastAPI/Django) • Next.js or React for the admin dashboard • PostgreSQL, schema design, migrations • REST API design, webhooks, JWT auth • Git, CI/CD, deployment on AWS / GCP / Vercel / Railway • Clean code, error handling, logging — production discipline AI / LLM (must-have) • OpenAI, Claude, or Gemini API — including structured outputs and function calling • Prompt engineering with reliability (retries, validation, guardrails) • RAG implementation with pgvector or similar • Bonus: LangChain, LlamaIndex Automation (must-have) • Make.com, n8n, or Zapier at an advanced level (custom modules, code steps) • HubSpot API and workflow customization • Comfortable choosing when to use no-code vs. custom code Bonus • Bland.AI or Vapi for voice automation • Apollo / Clearbit enrichment APIs • Looker Studio / GA4 reporting • Sentry, Datadog, or similar observability tools Ideal Freelancer / Agency • 3+ years shipping production software (not just automation demos) • Has delivered at least 2 similar AI + custom-dev projects — share GitHub repos, live URLs, or Loom demos • Comfortable making architecture decisions and defending them • Documents as they go — no black boxes • Available for 2 sync calls per week Screening Questions Please answer the following in your proposal: • Share 1–2 projects where you built both a custom backend/dashboard and the AI automation layer on top. Link to code or demo. • Walk us through your proposed architecture for Part A in 4–5 bullets — what stack, what hosting, and why. • How do you handle LLM unreliability in production (malformed outputs, rate limits, hallucinations)? • Which parts of Part B would you keep in no-code (Make/n8n) vs. move into custom code, and why? • Are you an individual or agency? If agency, who exactly writes the code and who manages the project? About Us 12-person B2B services company. This is Phase 1 of a larger automation and internal-tooling roadmap — if this project goes well, we plan to move to a monthly retainer ($2K–$5K/month) for ongoing development, AI feature work, and new workflows. Please do not send generic proposals. We shortlist based on specific examples, thoughtful answers, and evidence you can do both the hardcore dev and the automation side.