Sr AI/Agentic Engineer

B.S.D. Capital Inc. dba Lendistry; Los Angeles (CA); On-site; USD 111,200 - 185,000; Full time

Benefits offered by this job

Comprehensive Medical, Dental and Vision Insurance

Generous Paid Time Off

401(k) Match

Professional Development Courses

In-Office Provided Snacks and Drinks

Gym Facilities (LA & Tustin)

Job summary

B.S.D. Capital Inc. dba Lendistry is looking for an AI Software Engineer to lead the development of innovative AI applications. The candidate will work on document intelligence pipelines and borrower-facing tools while ensuring high-stakes financial decision requirements are met.

This role requires strong software engineering experience, particularly with LLMs and Python. Benefits include comprehensive medical insurance and generous paid time off.

Qualifications

• 5+ years of software engineering experience, 3+ years building LLM applications.
• Expert-level Python for production systems.
• Production experience with unstructured data.

Responsibilities

• Lead delivery of document intelligence pipelines.
• Develop borrower-facing conversational AI.
• Design evaluation frameworks for AI systems.

Skills

Software engineering experience

Python

LLM applications

RAG systems

Agentic workflows

Cloud deployment

Education

B.S. or M.S. in Computer Science

Tools

AWS Bedrock

LangChain

Docker

Job description

Lendistry is an Equal Opportunity/Affirmative Action Employer. We consider applicants without regard to race, color, religion, age, national origin, ancestry, ethnicity, gender, gender identity, gender expression, sexual orientation, marital status, veteran status, disability, genetic information, or membership in any other group protected by federal, state, or local law. If you need assistance or accommodation due to a disability, you may contact us at [e-mail]. Lendistry does not accept unsolicited resumes from recruiters, employment agencies, or staffing firms.

What You’ll Be Doing

• Lead delivery of document intelligence pipelines that read loan applications, tax returns, bank statements and financial statements with human-level comprehension and audit trails.
• Build underwriting copilots that surface risk signals, policy checks and recommended conditions in real time for Lendistry underwriters.
• Develop borrower-facing conversational AI that helps small business owners navigate applications, understand decisions and manage their loans.
• Harden the shared AI platform – prompt registry, tool‑calling framework, evaluation harness, retrieval infrastructure and inference routing layer used by every product team.
• Define and ship the evaluation and observability layer that turns AI reliability into a measurable, managed property of the system.
• Own end‑to‑end LLM features – requirements, design, implementation, evaluation, deployment, production operation – across origination, underwriting, servicing and borrower experience.
• Design new agentic workflows: LLMs that plan, call tools, evaluate results and iterate across complex lending tasks with appropriate human‑in‑the‑loop controls.
• Maintain, debug and improve existing LLM‑powered features in production: prompt pipelines, retrieval systems, document intelligence stack.
• Fine‑tune foundation models (including LLaMA‑family open‑weight and Bedrock‑hosted models) using LoRA, QLoRA, instruction tuning and prompt optimisation to meet Lendistry‑specific tasks.
• Build RAG systems end‑to‑end – chunking, embedding selection, vector retrieval, hybrid search and re‑ranking tuned for financial documents and lending policy.
• Lead development of document processing pipelines that extract structured data from PDFs, scanned images and unstructured financial documents using OCR, layout understanding and LLM‑based extraction.
• Design validation, confidence scoring and fallback mechanisms that make AI outputs safe for regulated, high‑stakes financial decisions, with clear audit trails and escalation paths.
• Diagnose and resolve agentic failure modes – non‑determinism, prompt sensitivity, tool misuse, looping, context‑window exhaustion and retrieval gaps – and build patterns to prevent recurrence.
• Contribute to and shape the shared AI platform – the prompt registry, tool‑calling framework, evaluation harness, retrieval infrastructure, and inference routing layer owned by the AI team.
• Design evaluation frameworks that measure model quality, output reliability, retrieval accuracy and regressions across iterations – golden sets, LLM‑as‑judge scoring, human‑review harnesses.
• Instrument AI systems with observability (logging, metrics, traces, token and cost accounting, drift monitoring, alerting) and manage cost and latency at the feature level – token budgeting, response caching, model‑tier routing and batching strategies.
• Collaborate with product, credit, underwriting and platform engineering to translate business requirements into reliable LLM system designs.
• Mentor junior AI engineers through design reviews, code reviews and pairing.
• Lead proof‑of‑concept work to validate new AI use cases quickly, measure real business impact and scale what works into production.
• Use AI coding assistants (Claude Code, GitHub Copilot, Cursor or equivalents) as a standard part of the development loop for code generation, refactoring, testing, documentation and review.
• Follow human‑review process: maintain clear judgment and established criteria for when to trust, verify or override AI‑generated suggestions consistent with Lendistry’s AI policies and regulatory requirements.
• Adopt and share emerging agentic development tools across Lendistry engineering and facilitate use of agentic development concepts: multi‑step task automation, LLM tool use, prompt engineering for code generation, integration of AI agents into engineering workflows.

Qualifications

• 5+ years of software engineering experience, 3+ years building and shipping LLM‑powered applications in production.
• Expert‑level Python for production systems – clean architecture, type‑safe data modelling (Pydantic or equivalent), async patterns, testable design.
• Deep hands‑on production experience with at least one major LLM provider – AWS Bedrock, Anthropic Claude, OpenAI GPT, Google Gemini or equivalent – including tool/function calling, structured output and streaming.
• Proven track record designing and operating RAG systems end‑to‑end – chunking, embeddings, vector databases (Qdrant, Pinecone, Weaviate, OpenSearch or pgvector), retrieval and re‑ranking – including measurement and improvement of retrieval quality.
• Experience leading agentic workflows in production – LLM agents that call tools, reason across multiple steps, and autonomously complete multi‑stage tasks with appropriate safeguards and audit trails.
• Hands‑on experience with fine‑tuning and adaptation – LoRA, QLoRA, instruction tuning or preference tuning – and rigorous evaluation of model outputs rather than demo‑driven validation.
• Strong LLM tooling fluency – LangChain or LangGraph, LlamaIndex, DSPy, Hugging Face – with judgment to pick the right tool and willingness to build custom solutions when required.
• Production experience with unstructured data – extracting, classifying, and generating structured outputs from text‑heavy inputs including documents, forms and scanned images.
• Cloud and deployment depth – AWS preferred (including Bedrock), containerisation (Docker), and self‑hosted LLM serving (vLLM, TGI, Ollama or similar).
• Evaluation discipline – ability to design evaluation frameworks for non‑deterministic systems, build golden sets, and reason about output quality at scale.
• Strong debugging instincts for LLM‑specific failure modes – hallucinations, retrieval gaps, prompt drift, latency spikes and cost regressions.
• API and service design experience – exposing AI capabilities as reliable internal APIs with clear contracts, error handling and cost controls.
• Security & regulated‑industry awareness – working knowledge of LLM security concerns (prompt injection, data exfiltration, output filtering, secure inference for sensitive workloads) and discipline around PII and sensitive financial data (detection, redaction, minimisation).

Preferred Qualifications

• Experience in fintech, lending, banking, healthcare or another regulated or data‑sensitive industry.
• Experience fine‑tuning LLaMA or similar open‑weight models on domain‑specific corpora.
• Familiarity with document‑understanding models (LayoutLM, Donut, Nougat) and modern OCR tooling (Textract, Tesseract or equivalents).
• Background in NLP tasks such as named entity recognition, classification or semantic similarity.
• Experience building and operating shared AI platforms (prompt registry, evaluation harness, routing layer) consumed by multiple product teams.
• Experience mentoring engineers and leading design reviews.
• B.S. or M.S. in Computer Science, Machine Learning or equivalent experience.

Benefits

• Comprehensive Medical, Dental and Vision Insurance
• Generous Paid Time Off, Birthday Day Off, 12 Paid Company Holidays
• 401(k) Match
• FSA and HSA
• Paid Life Insurance, Paid Disability Insurance, Pet Insurance
• Employee Assistance Program (EAP)
• Professional Development Courses
• In‑Office Provided Snacks and Drinks
• Gym Facilities (LA & Tustin/CEC Offices)
• In‑Office Engagement Activities

Compensation Range

The US base salary range for this full‑time position is $111,200 – $185,000 annually. The range displayed on each job posting reflects the minimum and maximum base salary for new hires for the position across all US locations. Within the range, individual pay is determined by multiple factors such as job‑related skills, experience and state of residence. Variable compensation elements are not included in the listed base salary.

Physical Requirements

This is a stationary position that requires frequent sitting (approximately 95%), repetitive wrist motions, grasping, speaking, listening, close vision and the ability to adjust focus. It may also require occasional standing, lifting, carrying of 20 lbs or less, walking, kneeling, bending/stooping, twisting, pulling/pushing, and reaching above the shoulder. Employees must be physically able to perform the essential functions of the position efficiently.