Senior AI Engineer - Political Intelligence & Agentic Systems

Senior AI Engineer — Political Intelligence & Agentic Systems - LATAM remote · Part-time · Reports directly to the CEO

The role in one line: Architect the agentic brain behind an AI political war room used by presidents, parties, and mayors across Latin America to govern with real-time data.

What you'll build

• Agentic orchestration: planners, executors, validators, long-term memory, tool integrations (Firecrawl, Perplexity, Gemini, GPT-5, OCR, SQL), queues, retries, token & credit budgets.

• Deterministic workflows on non-deterministic LLMs: scraping → structured extraction → analysis → publication, with full audit trail (which model, which source, what confidence, what cost).

• Political memory & RAG: ontology of parties, legislators, bills, votes, polls, social posts; embeddings; citation-enforced retrieval; verifiable anti-hallucination.

• Real-time alert system: detect emerging crises, sentiment spikes, opposition statements, surprise votes — push to the client before the press.

• Electoral prediction engine: combine official electoral results (2020/2024) + poll-of-polls + regional sentiment + legislative affinity → live "if the election were today" map.

• Honest calibration: every output carries a confidence score; no evidence → "no data". We never fabricate.

• Cost as a first-class citizen: you own driving monthly LLM + scraping spend down 50% without quality loss.

Stack TypeScript · TanStack Start · Supabase (Postgres + RLS + pg_cron) · Firecrawl v2 · Perplexity Sonar · Gemini 2.5/3.1 · GPT-5 · Cloudflare Workers

Must-haves

• 5+ yrs production systems, 2+ yrs production LLM systems (not demos, not notebooks).

• Real agentic orchestration: tool-use, planning, persistent memory, error recovery, observability. You know the trade-offs between LangGraph, custom frameworks, and explicit state machines.

• LLM evaluation: golden sets, regression, prompt A/B, hallucination measurement with verifiable citations.

• Senior TypeScript. Comfortable SQL (CTEs, windows, indexes). Postgres + pg_cron. • Serious scraping: Firecrawl, Playwright, anti-bot, rate limits, per-endpoint budgets.

• Genuine interest in politics and government. You read the news, you understand how a Congress works, you know quota vs probabilistic polling, you can tell a caucus from a bloc, and the idea that your code influences a presidential decision excites you.

Nice-to-haves

• Electoral data, polls, ballot OCR, multilingual sentiment (Spanish with regional slang).

• Data journalism, civic tech, campaign tech (NGP VAN, Civis, BlueLabs), public affairs.

• You've worked with political scientists, pollsters, or strategists.

Red flags, we'll pass

• "LLM engineer" who's only built RAG chatbots over PDFs.

• Zero interest in politics — this isn't a side project, it's the product.

• You think more prompting fixes everything. Here architecture > prompts.

How we work Report directly to the CEO. Fast decisions, no committees. Real client in production from day 1. Partner-level culture: you own the technical brain. Short meetings, heavy async, docs > meetings.

Compensation Competitive regional salary + early-stage equity + compute budget — no permission needed to experiment.

Apply here >> https://konsultora.ai/apply/palacio-analyst