Agentic AI Engineer to help optimize our media monitoring stack
Budget: $15 – $25 USD
Agentic AI Engineer — Optimize Our Media Monitoring Stack (Policy / Public Affairs)
About the Role
We run a media monitoring operation in the public policy space and need an experienced agentic AI engineer to help us **optimize an existing stack** — not build from scratch. We already have a working system; we want to push its quality, precision, and coverage substantially higher.
You’ll be working with someone who knows this space and has strong opinions about agent architecture, retrieval strategies, and prompt design. We’re looking for a peer who can pressure-test our setup and make it noticeably better.
What We’re Monitoring
- **Approximately 600-800 keywords** — policy terms, organizations, named entities, legislation, and topical phrases
- **Approximately 600 media sources** spanning:
- National and regional news outlets (online editions)
- Policy-focused blogs and trade publications
- Radio show and podcast websites (show pages, transcripts, episode descriptions)
- Social platforms: X (Twitter), Instagram, Facebook, YouTube, LinkedIn
- **Search modalities:** Boolean operators, proximity syntax, wildcards/stemming, full-text search, and semantic/vector search
The end goal is high-quality, low-noise capture of stories and posts that genuinely matter in the policy conversation — not a firehose of loosely-related hits.
Current Stack
We’re already running:
- **OpenClaw** as our agentic orchestration layer (Telegram-based command surface)
- **Claude Sonnet** (via OpenRouter) and **OpenAI Codex / GPT-5.4** as the model backbone
- **Exa** for semantic search and retrieval
- **Tavily** for web search (currently being phased out — your input on the replacement is welcome)
- General web search tooling
- Specialist sub-agents handling RSS ingestion, X/Twitter monitoring, and database persistence (PostgreSQL)
If you’ve worked with **OpenClaw**, **Hermes**, or similar agentic frameworks (LangGraph, CrewAI, AutoGen, custom orchestration over Anthropic/OpenAI APIs), say so explicitly in your proposal.
What We Need You to Do
The work is optimization, not greenfield construction:
1. **Improve keyword search precision** — design Boolean/proximity/stemming patterns at scale across 600 keywords; reduce false positives without losing real signal
1. **Strengthen semantic retrieval** — tune Exa usage (and/or recommend complementary tools), improve query formulation, evaluate hybrid keyword + semantic approaches
1. **Increase source quality and coverage** — audit our 600 sources, identify gaps in policy-relevant outlets, and improve ingestion reliability
1. **Optimize agent orchestration** — review how OpenClaw routes work between specialist agents and models; reduce token waste, improve handoffs, tighten prompt design
1. **Improve social media capture** — particularly X, YouTube, and LinkedIn, where access models are constrained and require thoughtful workarounds
1. **Build evaluation loops** — relevance scoring, dedup quality, source reliability tracking, so we can measure whether changes are actually improving output
Required Experience
- Hands-on experience building or operating **agentic AI systems** in production (orchestration, tool routing, multi-agent handoffs)
- Strong working knowledge of **retrieval architectures** — hybrid search, RAG, semantic + keyword fusion, reranking
- Deep familiarity with **Claude (Anthropic)** and **OpenAI** APIs, including prompt engineering for retrieval and structured outputs
- Experience with **Exa**, **Tavily**, **Brave Search**, or comparable search/retrieval APIs
- Working knowledge of **social platform APIs and their constraints** — especially X API v2, YouTube Data API, Meta Graph API, and the realities of LinkedIn access
- Comfort with **PostgreSQL** and structured data pipelines
- Ability to write tight, modular prompts and reason about token economics
Nice to Have
- Background in **media monitoring, public affairs, or policy intelligence**
- Experience with **TwitterAPI.io** or similar third-party social ingestion services
- NLP work in **named entity recognition, topic clustering, or relevance scoring**
- Familiarity with **Telegram bot integration** as an operational surface
- Multi-language monitoring experience
Important — Please Read Before Applying
We’ve already done a fair amount of work on this stack and have specific architectural opinions (small prompts, retrieval-by-reference over giant context, structured systems as authoritative sources, durable notes). We’re not looking for someone to rebuild from scratch or sell us a different product.
We’re also realistic about platform constraints: LinkedIn monitoring is structurally limited, X requires paid API access, Meta has tight scraping rules. **Tell us how you’d actually handle each platform** rather than promising blanket coverage.
To Apply, Please Include
1. **Specific examples** of agentic AI systems you’ve built or optimized — ideally with retrieval, search, or monitoring components
1. **Your read on our stack** — what jumps out as worth optimizing first, what you’d question, what you’d replace
1. **Your approach to relevance optimization** at the scale of ~600 keywords × ~600 sources
1. **How you’d handle each social platform** given current API realities
1. **Rate** (hourly or project-based) and rough **time estimate** to meaningful improvements
1. **Any clarifying questions** about scope, KPIs, or current pain points
We’ll prioritize applicants who engage substantively with the stack and the platform-access realities over generic pitches. If your proposal could have been written without reading this post, we’ll skip it.
Looking forward to hearing from you!
About the Role
We run a media monitoring operation in the public policy space and need an experienced agentic AI engineer to help us **optimize an existing stack** — not build from scratch. We already have a working system; we want to push its quality, precision, and coverage substantially higher.
You’ll be working with someone who knows this space and has strong opinions about agent architecture, retrieval strategies, and prompt design. We’re looking for a peer who can pressure-test our setup and make it noticeably better.
What We’re Monitoring
- **Approximately 600-800 keywords** — policy terms, organizations, named entities, legislation, and topical phrases
- **Approximately 600 media sources** spanning:
- National and regional news outlets (online editions)
- Policy-focused blogs and trade publications
- Radio show and podcast websites (show pages, transcripts, episode descriptions)
- Social platforms: X (Twitter), Instagram, Facebook, YouTube, LinkedIn
- **Search modalities:** Boolean operators, proximity syntax, wildcards/stemming, full-text search, and semantic/vector search
The end goal is high-quality, low-noise capture of stories and posts that genuinely matter in the policy conversation — not a firehose of loosely-related hits.
Current Stack
We’re already running:
- **OpenClaw** as our agentic orchestration layer (Telegram-based command surface)
- **Claude Sonnet** (via OpenRouter) and **OpenAI Codex / GPT-5.4** as the model backbone
- **Exa** for semantic search and retrieval
- **Tavily** for web search (currently being phased out — your input on the replacement is welcome)
- General web search tooling
- Specialist sub-agents handling RSS ingestion, X/Twitter monitoring, and database persistence (PostgreSQL)
If you’ve worked with **OpenClaw**, **Hermes**, or similar agentic frameworks (LangGraph, CrewAI, AutoGen, custom orchestration over Anthropic/OpenAI APIs), say so explicitly in your proposal.
What We Need You to Do
The work is optimization, not greenfield construction:
1. **Improve keyword search precision** — design Boolean/proximity/stemming patterns at scale across 600 keywords; reduce false positives without losing real signal
1. **Strengthen semantic retrieval** — tune Exa usage (and/or recommend complementary tools), improve query formulation, evaluate hybrid keyword + semantic approaches
1. **Increase source quality and coverage** — audit our 600 sources, identify gaps in policy-relevant outlets, and improve ingestion reliability
1. **Optimize agent orchestration** — review how OpenClaw routes work between specialist agents and models; reduce token waste, improve handoffs, tighten prompt design
1. **Improve social media capture** — particularly X, YouTube, and LinkedIn, where access models are constrained and require thoughtful workarounds
1. **Build evaluation loops** — relevance scoring, dedup quality, source reliability tracking, so we can measure whether changes are actually improving output
Required Experience
- Hands-on experience building or operating **agentic AI systems** in production (orchestration, tool routing, multi-agent handoffs)
- Strong working knowledge of **retrieval architectures** — hybrid search, RAG, semantic + keyword fusion, reranking
- Deep familiarity with **Claude (Anthropic)** and **OpenAI** APIs, including prompt engineering for retrieval and structured outputs
- Experience with **Exa**, **Tavily**, **Brave Search**, or comparable search/retrieval APIs
- Working knowledge of **social platform APIs and their constraints** — especially X API v2, YouTube Data API, Meta Graph API, and the realities of LinkedIn access
- Comfort with **PostgreSQL** and structured data pipelines
- Ability to write tight, modular prompts and reason about token economics
Nice to Have
- Background in **media monitoring, public affairs, or policy intelligence**
- Experience with **TwitterAPI.io** or similar third-party social ingestion services
- NLP work in **named entity recognition, topic clustering, or relevance scoring**
- Familiarity with **Telegram bot integration** as an operational surface
- Multi-language monitoring experience
Important — Please Read Before Applying
We’ve already done a fair amount of work on this stack and have specific architectural opinions (small prompts, retrieval-by-reference over giant context, structured systems as authoritative sources, durable notes). We’re not looking for someone to rebuild from scratch or sell us a different product.
We’re also realistic about platform constraints: LinkedIn monitoring is structurally limited, X requires paid API access, Meta has tight scraping rules. **Tell us how you’d actually handle each platform** rather than promising blanket coverage.
To Apply, Please Include
1. **Specific examples** of agentic AI systems you’ve built or optimized — ideally with retrieval, search, or monitoring components
1. **Your read on our stack** — what jumps out as worth optimizing first, what you’d question, what you’d replace
1. **Your approach to relevance optimization** at the scale of ~600 keywords × ~600 sources
1. **How you’d handle each social platform** given current API realities
1. **Rate** (hourly or project-based) and rough **time estimate** to meaningful improvements
1. **Any clarifying questions** about scope, KPIs, or current pain points
We’ll prioritize applicants who engage substantively with the stack and the platform-access realities over generic pitches. If your proposal could have been written without reading this post, we’ll skip it.
Looking forward to hearing from you!