AI Software Engineer - LLM Applications
First seen by Alion on Sep 9, 2026.
About Top of Mind Labs :
Top of Mind Labs builds AI systems for private equity firms, their portfolio companies, and mid-market businesses. Our clients are investment firms, wealth-management platforms, legal technology providers, and software companies who need production AI that works on their actual data and workflows.
Our mission is to give every company a brain that remembers everything and uses it. In practice, that means building harnesses for knowledge workers - the lawyers, accountants, analysts, and engineers who do the highest-value work inside these organizations - so that the systems they use hold context, retrieve it accurately, and act on it.
We are a small, senior team. Engineers here work directly with client stakeholders, own systems end to end, and ship to production inside four - to - six week engagement cycles.
What you will build :
- Agentic assistants embedded in business surfaces. Agents that live in Slack and Microsoft 365, read across email, documents, and internal systems, and take action - drafting, reviewing, scheduling, and preparing work product for the human in the loop.
- Document intelligence pipelines. Extracting structured, verified data from dense legal and financial documents, with confidence scoring and human review workflows where accuracy is non-negotiable.
- Retrieval systems over domain corpora. RAG applications for tax, legal, and financial-analysis use cases, where a wrong or ungrounded answer is worse than no answer.
- Semantic layers and knowledge graphs. Ontology design and entity resolution over enterprise data so that downstream agents can reason across systems rather than one database at a time.
- Full applications. Internal platforms that replace spreadsheets and manual processes - data model, API, front end, analytics, and deployment.
You will not be training foundation models. You will be building reliable systems on top of them.
Core requirements :
- 4+ years building and shipping production software, with at least 1 - 2 years on LLM-based applications
- Strong Python. Working proficiency in TypeScript
- Demonstrated experience building LLM applications in production - not prototypes, not notebooks. You have dealt with latency, cost, failure modes, and users who found the edge cases
- RAG systems end to end : chunking strategy, embedding selection, hybrid search, reranking, grounding and citation
- Structured output and tool use : JSON schema enforcement, function calling, validation and repair loops
- Agent development using LangGraph, LangChain, or an equivalent framework - tool use, multi-step planning, state and memory management
- Solid SQL and data modeling; comfort with a modern warehouse (Snowflake, BigQuery, or similar)
- API design, authentication, background jobs and queues
- Enough infrastructure ability to deploy and operate what you build : Docker, CI/CD, logging, cost and latency monitoring
Strongly preferred :
- Evaluation experience - you have built test sets, LLM-as-judge pipelines, or regression suites for non-deterministic systems, and can explain how you knew a change was an improvement
- Integration work against Slack APIs, Microsoft Graph, or Google Workspace APIs
- MCP (Model Context Protocol) server or client development
- Document extraction from complex PDFs : OCR, vision-language models, layout-aware parsing
- Knowledge graph or ontology work - RDF/OWL, property graphs, Neo4j, entity resolution
- React or Next.js
- Open-source contributions to agent frameworks, inference tooling, or LLM libraries. This is the single strongest signal on a resume for this role - include links
- Exposure to open-weight models and self-hosted inference (vLLM, Ollama)
- Domain background in financial services, legal, accounting, or investment workflows
How you work :
- Client-facing communication. English fluent in writing and on calls. You will join client working sessions, demo your own work, and explain technical tradeoffs to non-technical stakeholders
- Comfortable with ambiguity. Requirements change mid-sprint. Scope arrives as a conversation, not a ticket
- Judgment over volume. We use AI tooling heavily across our own development. What we need from engineers is architectural judgment, taste, and the ability to tell when a system is actually working
Our hiring process :
- Application review - 2 - 3 days
- Intro call (30 min) - role, background, mutual fit
- Paid work sample (6 - 8 hours, compensated at market rate) - a real problem drawn from past engagement work. You keep the code
- Technical deep dive (60 min) - walkthrough of your work sample and your prior systems
- Final conversation with the founder (45 min)
- Offer
We aim to complete the full process within two weeks.
New hires begin with a paid four - to - six week trial engagement on live client work before converting to a permanent role.
Why this role :
- Direct ownership. Small team, no layers between you and the client problem
- Range. Agents, retrieval, extraction, knowledge graphs, and full-stack delivery - usually within the same quarter
- Compensation set against global rather than local benchmarks, for engineers who clear the bar
- Employment through an employer of record with full statutory benefits
Skills
Machine Learning, Python, LLM, Agentic AI, Artificial Intelligence, TypeScript