Financial information in Turkey moves fast, and most of it arrives in dense, inconsistent formats. Redar is my answer: an AI-assisted analysis system that tracks Kamuyu Aydınlatma Platformu (KAP) disclosures and other open sources, detects the important ones, and turns them into concise, useful summaries—delivered on the web and instantly via Telegram.
Why I Built Redar—and What It Does Differently
I started Redar in early September after a conversation with Onur Pıtır. He had a need, I dug in, scoped the requirements, and ended up building something far more advanced than the initial ask. The goal was clear: automatically surface the right disclosures and make them understandable the moment they land.
Redar focuses on Turkey’s markets and is currently a closed, private system we use ourselves. That decision is intentional. While the audience could logically include traders, funds, and finance teams, the cost of BIST data licensing—especially once we’d add product-level features—doesn’t fit our current priorities. If we partner with a brokerage (aracı kurum), we can take Redar public as a full product; Onur is exploring this path.
What sets Redar apart is the combination of speed, structure, and delivery. Many finance platforms exist, but none offer this specific blend: automatic KAP tracking, multi-source news ingestion, smart extraction and LLM-based summarization for messy formats, search and filtering across disclosures, and Telegram alerts in exactly the format we want. It’s built for actionable clarity, not dashboards for their own sake.
Under the Hood: From Unstructured KAP to Clean, Actionable Signals
KAP disclosures are not standardized and the HTML markup is often poor—that’s the core technical challenge. Redar tackles it in two ways:
- Deterministic parsing for any source that arrives in a standard structure.
- LLM-assisted parsing and summarization when structure breaks down, so we can still extract the key entities and meaning.
The system continuously pulls KAP and various news sources. When a disclosure we care about drops, Redar extracts the relevant information and renders a compact, readable summary on the web while pushing a Telegram notification instantly. We’ve leaned into Telegram heavily—part learning project, part pragmatic delivery—so there’s more capability there than strictly necessary, spanning both Bot and WebApp features.
Tech stack at a glance
- Backend: Fastify
- Frontend: Nuxt v4
- Data: MariaDB
- Caching: Redis
- LLM: OpenAI API (different models for different tasks—e.g., Nano, Mini—no custom training yet)
- Networking/Infra: Traefik reverse proxy, Docker Swarm on my existing VPS
- CI/CD: GitHub Actions
- Search & Ops: Server-side filtering and querying with latency kept tight through Redis and careful parsing paths
This stack lets us ingest quickly, normalize as needed, and respond in near real time. When structure is present, we parse deterministically for speed and stability. When it’s not, we defer to the LLM to summarize and extract fields so users still receive coherent, comparable outputs.
Constraints, Lessons, and What’s Next
Licensing and access. BIST data licensing is expensive. Even if the current private setup doesn’t require it, the moment you productize with richer market-level features, costs jump. That’s why Redar is invite-free and private for now. With the right brokerage partnership, we could justify the leap and bring a public product to market.
Parsing the unparseable. The lack of KAP standardization forced us to blend rules and models. The lesson: treat LLMs as fallback structure builders—great at cleaning up chaos—while preserving deterministic paths wherever the source allows it.
Telegram as a first-class channel. Over-investing in Telegram wasn’t an accident. Alerts beat dashboards when time matters, and the exact-format messages we needed were trivial to trigger once the plumbing was in place.
Roadmap. If the brokerage path materializes, we’ll scale Redar into a product-grade system with broader features and licensing compliance. That unlocks exposure to the wider audience we originally envisioned across traders, funds, and finance teams in Turkey.
If you’d like to talk partnerships or explore how Redar could evolve into a public product, reach out on X: @evrenbal and @onurpitir001. For an overview of the project, visit redar.com.tr.
