Senior LLM Engineer – Founding Team Offline

Altss is building the most advanced intelligence engine for private markets — turning millions of startup, fund, and allocator pages into structured investor insights using LLM enrichment and real-time data pipelines.

We’ve already parsed and profiled over 9,000 family offices. Now we need a founding-level LLM engineer to lead our enrichment and model infrastructure: LangChain chains, embedding pipelines, RAG systems, fine-tuning, and evaluation logic.

You’ll work side-by-side with our infra engineer, QA lead, and founder to deploy enrichment systems that outperform anything built by PitchBook, Harmonic, or Fintrx — and you’ll ship 10x faster with 1/10th the team size.

What You’ll Own

• Build and maintain modular LangChain pipelines for entity enrichment, mandate extraction, and principal profiling
• Design embedding + RAG systems for unstructured document enrichment (e.g., fund bios, firm pages, press)
• Integrate both OpenAI models (GPT-4) and open-source LLMs (Mistral, Zephyr, LLaMA 3)
• Fine-tune OSS models using Axolotl, QLoRA, or PEFT
• Deploy and scale in-house inference using vLLM or TGI
• Implement confidence scoring, prompt evaluation, and hallucination detection
• Build prompt eval frameworks (LangChain eval, Promptfoo, GPT-as-judge)
• Work closely with QA to route low-confidence outputs and retrain weak chains
• Shape the full enrichment layer powering our LP and startup intelligence system

Stack You’ll Work With

• LangChain, OpenAI, Mistral, Claude, 
• Qdrant, Weaviate, FAISS (vector DBs)
• vLLM, TGI, Axolotl, HuggingFace Transformers
• Prefect 2.0, Postgres, DuckDB
• GPT eval chains, Promptfoo, custom JSONL evaluation
• Collaborate with: Infra (Prefect + pipelines), QA (flagging + scoring), Parsing team (5M+ pages)

What We’re Looking For

• 5–10+ years in NLP, backend ML, or LLM pipeline engineering
• Deep experience with LangChain, embeddings, chunking strategies
• Built production RAG pipelines, not just toy LLM apps
• Hands-on with fine-tuning OSS models (Mistral, Zephyr, LLaMA) using Axolotl, PEFT, or Flash Attention
• Can confidently deploy and monitor LLMs on vLLM, TGI, or custom inference stacks
• Strong grasp of prompt evaluation, enrichment scoring, and debugging hallucinations
• Bonus: Familiar with investment data (VC, PE, LPs) — or eager to learn fast

Why Join Us

• You’ll build the brain of a product already beating legacy competitors
• You’ll ship weekly, not wait months
• You’ll work with a senior team that gets technical speed, quality, and independence
• We don’t do hype — we build real data infrastructure and sell it to real clients