National-level financial regulatory institution
Worked on AI application delivery in financial regulatory settings, with attention to security review, permission boundaries, auditability, and production stability.
About Nie Er
I deliver enterprise AI applications in high-compliance, sensitive-data environments.
I am Nie Er, a ten-year veteran software engineer. Since 2024, my work has focused on enterprise LLM application delivery, including RAG knowledge bases, complex document parsing, information extraction, AI support assistants, and agent workflows.
Instead of building demo-first prototypes, I am stronger at turning a company's existing documents, knowledge, and business processes into systems that can launch, run, be evaluated, be traced, and be maintained.
I have worked on projects for a national-level financial regulatory institution, one of China's leading securities firms, an A-share listed environmental group, and teams in healthcare and shipping logistics. These projects require more than model quality: they involve permission boundaries, data security, compliance review, audit trails, stability, and production delivery.
My core advantage is not merely connecting an LLM API. It is getting AI applications into real systems under strong compliance, sensitive data, and complex workflow constraints.
Client Scale
Sanitization should not hide the value of the work.
Public material is sanitized as needed: no direct customer names, internal data, or sensitive system details are disclosed. But the client scale, business problem, solution logic, engineering tradeoffs, and delivery outcomes remain visible.
One of China's leading securities firms
Worked across investment research, contracts, due diligence, and report automation scenarios where RAG, document parsing, and information pipelines had to fit real workflows.
A-share listed environmental group
Added AI capabilities to existing systems and operational workflows instead of building a separate demo-only interface.
Healthcare and shipping logistics
Handled projects in healthcare and shipping logistics, with emphasis on data boundaries, workflow design, and maintainable delivery.
Experience
The value is rarely in the model API alone. It is in the business boundary and delivery discipline.
Over the past few years, I have worked on AI application projects across financial regulation, securities research, healthcare, environmental utilities, and shipping logistics. The scenarios include investment research knowledge bases, private fund contract extraction, due diligence document parsing, macro analysis and asset allocation reports, internal knowledge assistants, AI support, and workflow automation.
Enterprise knowledge bases / RAG Q&A
Production-facing knowledge systems with document parsing, retrieval, reranking, cited answers, permission isolation, evaluation sets, and failure analysis.
Document parsing / information extraction
Structured extraction from contracts, due diligence files, research reports, announcements, and PDF tables, including batch processing, source evidence, human review, and version control.
Report and information pipelines
Stable pipelines for collection, cleaning, classification, summarization, review, and delivery across daily reports, weekly reports, policy briefs, industry reports, and internal reference notes.
AI upgrades for legacy systems
Adding Q&A, summarization, classification, natural-language query, and document extraction to existing OA, CRM, ERP, ticketing, knowledge base, or business systems.
Judgment
I usually start by deciding whether the problem is worth building, then decide how to build it.
Many enterprise AI projects fail not because the model is too weak, but because the problem was never defined clearly. Some scenarios need RAG, some need rules and traditional system integration, some need workflow automation, and some are not ready for AI yet.
Common risks
- The business scenario is too broad and the target is unclear.
- The data quality is weak, or the documents are not ready for direct use.
- There is no evaluation standard beyond demo impressions.
- Permission, compliance, and audit boundaries are not designed early.
- The team focuses on model integration but ignores long-term maintenance cost.
- No one owns continuous iteration after launch.
Fit
Projects I am a good fit for
I focus on LLM application implementation. Compared with demo-only work, I am a better fit for turning existing enterprise documents, knowledge, and workflows into systems that can launch, run, and be maintained.
- Enterprise knowledge bases / RAG Q&A: searchable, cited, traceable, and permission-aware.
- Document parsing / information extraction: contracts, due diligence files, research reports, announcements, PDF tables, and other structured extraction work.
- Report and information pipelines: daily reports, weekly reports, recurring summaries, scheduled runs, review, and delivery.
- AI upgrades for existing systems: adding AI capabilities to OA, CRM, ERP, ticketing, or knowledge base systems.
- Second-vendor / technical partnership: AI module development, PoC work, and delivery support for software companies, outsourcing teams, and system integrators.
Not a Fit
Projects I usually do not take on
To avoid wasting time on both sides, I usually do not treat the following as primary cooperation directions.
- High-risk automated decisions such as stock recommendations, automated trading, direct medical diagnosis, or final legal conclusions.
- Large all-in-one AI platforms without clear requirements, data, owners, or acceptance criteria.
- Low-price tool setup work where the only goal is the cheapest simple bot or workflow.
- Core systems that require special qualifications, classified environments, or primary compliance responsibility.
Within a proper boundary, I can still participate as a technical consultant, module developer, or second-vendor partner. I do not personally assume primary responsibility that should belong to a qualified organization.
How to Start
Send the project context first. I will start with a technical judgment.
If you are not sure whether your scenario is suitable for LLMs, send the basic context first. I will help judge whether it needs RAG, rules, workflow automation, or an upgrade to the existing system.
No courses, no generic solution pushing, only concrete problems.
Useful context to prepare
- Industry
- Business problem
- Existing documents, data, or systems
- Expected output
- Whether existing-system integration is needed
- Internal use or customer-facing use
- Consulting, project delivery, part-time collaboration, or full-time opportunity
Connect
Public profiles and update channels.
These links cover open-source work, career background, long-form writing, engineering notes, and shorter updates.