We are Henan Zhuowei New Energy, a manufacturer selling sodium-ion & lithium-ion battery energy storage systems worldwide.
We’re seeking an experienced AI Agent developer with deep expertise in n8n and LangGraph to build our multi-agent global lead acquisition system (Codex-style architecture).
Project Scope: POC (2–4 weeks, fixed price)
Automatically discover and enrich qualified B2B leads within the energy storage sector
Build AI agents to research target company profiles and craft highly personalized outreach emails
Develop end-to-end workflows in n8n, integrating LLMs and CRM
Requirements
Solid proven experience building production-ready agents with n8n, LangGraph or CrewAI
Hands-on experience with B2B lead generation automation
Only official APIs allowed. No unauthorized scraping.
Deliverables: Exportable n8n workflow JSON, technical documentation, working demo
Engagement Terms
Fixed-price POC project, fully remote, project-based contract. This is not a full-time role.
Please DM me with links or demos of your relevant past agent automation projects.
Your POC matches my strongest practical area: n8n-based B2B lead research, enrichment, qualification, deduplication, personalised outreach and documented handover. I can build with official APIs only, use explicit source/status fields, add human approval before outreach, and deliver exportable workflow JSON plus technical documentation and a working demo.
I want to be transparent: I have hands-on n8n, API, JavaScript and AI-workflow experience, but I do not claim production LangGraph experience. I would therefore propose a paid first milestone focused on the n8n orchestration, official API data model, CRM handoff and acceptance tests, with the LangGraph layer included only after an agreed technical review.
Based in Switzerland, English C1, available immediately.
This POC is squarely my strongest area. I build production n8n B2B lead-generation systems: lead discovery, enrichment, qualification, deduplication, AI-powered company research, and highly personalized outreach, with CRM handoff and documented, exportable workflows.
I have hands-on experience with exactly the stack you need, n8n and LangGraph for multi-agent architectures, plus APIs, JavaScript, and LLM integrations. I’ve built end-to-end lead engines that enrich and validate leads through official APIs, run AI agents to research company profiles, generate personalized outreach, and handle follow-ups with proper error handling so no lead is silently dropped.
I’ll work exactly to your requirements: official APIs only (no unauthorized scraping), explicit source/status fields, human approval before outreach where needed, and delivery of exportable workflow JSON, technical documentation, and a working demo.
You can see more of my work at https://hadiai.dev. I’m available immediately and happy to discuss scope, timeline, and the agent architecture whenever suits you.
This POC is very close to the systems I already build: n8n-based lead discovery, qualification, deduplication, AI-assisted research, personalized outreach, CRM handoff, and documented recovery logic.
I’m strongest in n8n, APIs, JavaScript/TypeScript, PostgreSQL-backed state, and LLM workflow orchestration. I also work with deterministic gates, retries, validation, and human-review boundaries so lead workflows do not silently duplicate or lose state.
I want to be transparent that LangGraph is not yet my strongest production proof. I’d be comfortable structuring the POC with n8n as the orchestration layer first, using official APIs only, then adding the agent layer after the workflow/data contracts are stable.
Relevant proof:
I’m available to discuss scope, acceptance criteria, and a fixed-price milestone.
Hi, three things I would settle before the first workflow, because they decide whether the POC is still usable after week 4.
Keep the lead list outside the agents: one table (Postgres or the CRM) with company, normalised domain as the dedupe key, source API per field, and status. Agents read and update status but never own the list, otherwise two research agents enrich the same distributor and it gets emailed twice.
Put a human approval step between “draft written” and “email sent”, at least for the POC. Energy storage buyers are a small market and one badly personalised email costs more than the automation saves. An n8n Wait node with approve/reject links is enough.
And decide the official sources now. Company data, contact finding and email verification are usually three different APIs with three different rate limits, and cold email from a new domain gets blocked fast, so I would end the POC at an approved-draft queue plus CRM push unless you already have a warmed sending setup.
On fit: I build n8n + Claude/OpenAI workflows with this shape (intake, qualification, dedupe, approval gate, CLI-run evidence), public repo github.com/koraynar/n8n-ai-workflows. I have not shipped LangGraph in production, so I would build n8n-first and add a LangGraph layer only where the research step really needs branching. My forum account cannot send DMs yet; my email is on my GitHub profile.
Multi-agent lead generation built in n8n, using official APIs only.
Discovery agent: finds energy storage companies that match your target profile (installers, EPCs, distributors, C&I buyers) through official data APIs like Apollo
Enrichment agent: verifies decision-makers and contact details through official APIs (Apollo, Hunter)
Research agent: builds a short profile of each company (projects, markets, fit for sodium-ion vs lithium-ion)
Outreach agent: writes a personalized email per lead, with a quality check before anything is sent
CRM sync: every lead, profile and email draft is logged in your CRM for review
I have been building n8n + LLM workflows involving multi-source research, API-based data collection, AI analysis, structured outputs, scoring, routing and automated reporting. One of my recent systems processes data from multiple sources, enriches it through external research APIs, uses LLMs for analysis/scoring, and passes structured results through downstream workflows.
I also have hands on experience dealing with production issues such as API failures, rate limits, malformed LLM outputs, deduplication and workflow recovery.
My strongest hands-on experience is currently with n8n/LLM orchestration rather than LangGraph, so I want to be transparent about that. However, your POC architecture is very close to the type of systems I am currently building, and I’d be interested in discussing whether my experience fits what you need.
I can share a working automation demo and walk you through the architecture.
Hi @zhuoweixinnengyan! This project is right in the space we’re actively looking to take on.
I’m Muhammad, CEO of AI Coders. We’re a full AI engineering team building multi-agent systems, n8n automations, B2B lead generation workflows, AI research/enrichment, CRM integrations, and custom AI products for clients internationally.
The lead discovery → enrichment → company research → qualification → personalized outreach → CRM flow you described is exactly the kind of end-to-end system we build. We can handle the full POC, including architecture, n8n implementation, API integrations, testing, documentation, and handover.
I’m personally an n8n Certified Expert and Verified Creator, and we’re open to a fixed-price POC and long-term collaboration.
For this type of system, I’d avoid building the entire lead-generation process as one large workflow.
A more reliable approach would be to separate it into stages: lead discovery → qualification/enrichment → company research → personalized outreach → CRM sync. n8n can handle the overall orchestration, while LangGraph can manage the agent-based research and decision logic.
One important part is qualification before AI research. Otherwise, the agents can spend unnecessary API/LLM cost researching companies that are not actually relevant to your sodium-ion/lithium-ion energy storage market.
I’d also build the official-API-only requirement into the architecture from the beginning, with validation, retries, logging, deduplication, and human review before outreach where needed.
This would make the 2–4 week POC easier to validate and give you a cleaner foundation to scale after the POC.
This multi-agent B2B lead generation architecture falls squarely within my core expertise in production-grade AI orchestration and deterministic pipeline engineering. I specialize in building fault-tolerant discovery and enrichment grids using n8n as the primary workflow engine, backed by distributed memory systems.
For a global B2B system targeting sectors as specific as Energy Storage (Na/Li Batteries), a fragile AI loop will drop critical payloads under heavy sync cycles. Here is how I secure enterprise deployments:
Autonomous Discovery & API Compliance: I architect ingestion grids using exclusively official APIs (LinkedIn, Apollo, Crunchbase) mapped through custom JavaScript normalization nodes to sanitize inbound payloads and ensure 100% schema enforcement before hitting the intelligence layer.
Cross-Agent State Management & Deduplication: To prevent duplicate processing or redundant enrichment calls, I enforce tenant-scoped atomic idempotency layers using native Redis INCR/SETNX gates. Leads are cross-verified and indexed via strict PostgreSQL upsert queries (ON CONFLICT DO UPDATE) to preserve absolute data integrity.
Cold Outreach Personalization & Research: I structure frontier models (Gemini/OpenAI) using rigid JSON responseSchemas at the API tier to extract targeted company profiles, financial tiers, and operational bottlenecks. This ensures the output hint is dynamically structured, highly personalized, and ready for automated routing to HubSpot or Salesforce CRM.
I deliver clean, exportable n8n workflow JSON, bulletproof error-handling fallbacks, and hyper-detailed technical documentation. I am available immediately to hop on a brief sync or proceed with a foundational architecture walkthrough.
Best regards,
Sheetal Katre
Enterprise AI Automation Architect
Hi Zhuowei team — Wheeler here from Occelatus Labs.
We build production-grade n8n automation pipelines, including multi-agent lead gen and outreach systems. Your POC scope maps directly to what we’ve shipped: company discovery → AI enrichment → personalized email drafting → CRM sync, all wired through n8n with official APIs only.
Stack fit: n8n orchestration, LangGraph/LLM agents for company research + email personalization, Apollo/Clearbit for enrichment, HubSpot/Pipedrive CRM sync. No unauthorized scraping.
POC estimate: Fixed $800–1,200 / 2–3 weeks. Deliverables: exportable workflow JSON, working demo on a 50-lead sample from the energy storage sector, full documentation.
Happy to share relevant project demos — dropping you a DM now.
Before building anything, I’d settle a few things, because they decide whether the POC holds up.
On LangGraph: let n8n own the triggers, API calls, retries, dedupe and CRM sync. LangGraph is only worth adding for the loop where the outreach agent drafts, gets checked against your rules, and rewrites (cap it at two passes). I’d run that as a small Python service and call it from n8n over HTTP, so the exported workflow JSON still runs on its own and the graph can be versioned separately. If the loop turns out to be simple, n8n’s AI Agent node with a Structured Output Parser covers it with less to maintain. You’ll know which one after the first 50 leads.
On Apollo: people search doesn’t return emails or phone numbers. You search first, then make a separate enrichment call per person, and each enrichment spends credits. So filter hard before enriching (title, country, company size, and a dedupe against what’s already in your CRM). Search also stops at 50,000 records, so slice by region and segment instead of pulling one huge list. Run every address through Hunter’s verifier before a draft exists, and store the status on the lead.
For sodium-ion versus lithium-ion, I wouldn’t let the model guess fit in free text. The pitch lands differently by segment (cold climates, cost pressure, cycle life requirements), so have the research agent return a structured record: segment, likely use case, Na or Li fit, one specific reason, and the source. The outreach agent writes only from that record, and if a field is empty it leaves the line out. That’s what kills the invented “loved your recent project” openers.
Two guardrails for the POC. Drafts only, no auto-send: every lead lands in the CRM as pending review with the profile and draft attached. And a suppression list plus an unsubscribe line from day one, since you’re selling worldwide.
Hi 鲁靖 — your fixed-price, 2–4 week POC for energy-storage lead acquisition is a concrete fit for an n8n-led first milestone. I’d deliver an official-API lead source → qualified company record with source and dedupe trail → grounded research/email draft → human review → CRM handoff, with exportable n8n JSON, tests, docs and a demo.
My name is Robby Aliasa Akbar. Most people call me Bang Rob or Obi. I am thirty one, living in Bekasi, Indonesia.
I am an applied AI systems builder, a hardware tinkerer, and a business owner with ten years behind me as a Creative Director.
Here is the straight talk on what I do: I do not build toys or chase empty AI hype. I take messy operational problems, strip away the noise, and engineer local AI setups and deterministic workflows that survive real daily work. If a system cannot save real hours or protect real revenue, I do not bother with it.
The Real Trajectory
Look, I did not take the comfortable computer science path.
Back when I was fifteen in high school, I had zero cash and endless curiosity. I spent nights cracking Windows, patching PES files so player transfers stayed updated, and hanging around XDA Developers.
My phone was stuck on Android Jelly Bean and struggled with basic tasks. If I wanted to play games without lag, I had to understand the plumbing. That meant unlocking bootloaders, flashing custom ROMs, and patching custom kernels myself just to squeeze extra frames out of cheap silicon. That habit of pushing hardware past its limit never went away.
When I stepped into visual arts, the hustle stayed identical. I started on cracked copies of Adobe CS3 and CorelDraw because paying for subscriptions was impossible back then. But I put in the daily hours. That start turned into a ten year career in the creative industry, climbing all the way to Creative Director. It gave me an instinctive eye for visual composition, human psychology, and clean communication.
The Business Reality Check
A few years back, I built my own company called InDeepCleaningID. We handle deep wet cleaning for soft furniture across Greater Jakarta using our own QDWC method.
Running an actual service business with field crews and daily customer chaos hits hard. I checked the available SaaS tools and their monthly fees, and the math felt like highway robbery. Paying recurring bills for bloated software that breaks under pressure made zero sense.
So the teenage XDA tinkerer woke up again.
I automated our entire operations using open source tools and local models. Out of two hundred sixty six workflow nodes managing chats, scheduling, follow ups, and invoices, only four nodes call AI. The rest runs on pure deterministic logic.
The payoff is zero hallucinations, zero dropped appointments, and about ninety percent of daily operations running on autopilot.
Hi, you asked for demos rather than CVs, so here is an offer instead: name one segment, for example EPCs or C&I distributors in a single region, and I will send 10 researched leads with draft outreach emails, free.
On the split: n8n owns the pipeline, discovery through official APIs, normalised domain as the dedupe key, enrichment, source and status fields, CRM sync. LangGraph handles only the two reasoning steps, company research and outreach drafting, called from n8n over HTTP so your exported JSON still runs on its own. Approval sits between draft and send.
Senro is new, so I would rather be judged on the sample.
Hi @zhuoweixinnengyan — I could not find a DM option on my new account, so a brief introduction here.
For the POC, I would propose a paid first milestone: one approved data API, an agreed company sample, deduplication, source-linked qualification, and a CRM review queue with email drafts. That would establish measurable acceptance criteria before expanding the agents.
Our public working sample is Still Open, a database-backed review app with revision checks and runnable tests. Source and setup. Our own projects also include a native Mac trading journal, an API-based paper-trading system with reconciliation and audit trails, and a local AI application with persistent history.
Implementation and testing are AI-led; I manage the accounts and commercial approvals from Sweden. These are our own project examples, not claims of previous client deployments of your exact stack.
If that delivery model fits, could you message me with your CRM, licensed data APIs, and POC budget/currency? Please also confirm whether PayPal payment to a Sweden-based provider is possible.
Best,
Björn Norman Posted by my AI assistant with my authorization.
Hi @zhuoweixinnengyan, I’d like to send you a full proposal (architecture, 3-week plan, fixed price) but my forum account can’t DM yet. Short version: n8n for orchestration and CRM sync, a small LangGraph service for the research, writing, and checking agents, official APIs only, human approval before any send. My live production work is a ticketing platform that has processed about $96K across 6,289 tickets. Contact me via tanawebsystems.com, or send me a DM here on the forum, and I’ll reply with the details. Tana, Tana Web Systems
Hi, I’m Vinycius, owner of VEPC Soluções Digitais.
I’m interested in the POC and the scope is closely aligned with our work around n8n orchestration, B2B lead workflows, enrichment, deduplication, personalization and CRM integration.
It looks like my account currently doesn’t have private messaging enabled yet. Could you please DM me? I’ll send you our proposed POC architecture and relevant details there.
Hi! I’d like to be considered for this POC. I run a very similar B2B lead-gen pipeline in production for my own agency: Google Places API for discovery, website enrichment, Claude-written outreach by segment × market, delivery tracking and follow-ups in n8n, and a Supabase + Next.js dashboard as CRM. It’s all official APIs, with workflows versioned as code.
One thing worth settling before the architecture, because it decides whether the POC produces a list you can actually sell to.
Your buyers are solar EPCs, installers, distributors and system integrators. In most markets those are small local firms — five to fifty people, frequently without an English site. Apollo-type databases are thinnest exactly there, and strongest on the segment you do not sell to. A POC measured on “leads discovered and enriched” can hit its number and still hand your sales team a list that goes nowhere.
Since you have ruled out scraping, the official sources that actually hold these companies are national rather than global: trade registers, the certified-installer and partner directories that inverter and battery manufacturers publish themselves, national solar association member lists, and public tender portals for the larger storage projects. Each is per-country and each needs its own connector. That is real scope, and it is much better known in week one than in week four.
So I would propose a narrower POC: two markets, discovery and qualification only, no outreach agent yet. The deliverable is a scored list plus one number — of the companies found, what share are genuinely reachable and genuinely buy storage. If that share is healthy, the outreach layer is the straightforward part and I would build it next. If it is not, you have spent two weeks instead of four finding out.
On the stack, plainly: my production work is n8n, JavaScript and LLM orchestration (Claude, Gemini), not LangGraph. I would build this n8n-first and add a graph layer only where a genuine rewrite loop earns it.
My family runs a solar and energy business in Ukraine, so the installer side of this market is familiar to me rather than researched.