Petar Taneski / Software engineer

Software that works.
Details that

I build software that turns complex tasks into usable tools, from dashboards and product interfaces to data pipelines and automation.

Illustration of Petar working at his laptop

Full-stack systems

Every number has a trail.

An equity research tool co-built with Marija Maneva. Type a ticker and get a first-draft report in about 15 seconds, where filing figures link to SEC and EDGAR and calculated figures show their formula and inputs.

Sources
  • SEC EDGAR filings
  • FMP market data
  • FRED Treasury yield
  • Claude
FastAPI backend
  1. Resolve the ticker
  2. Fetch every source in parallel
  3. Extract the financial statements
  4. Ratios and a DCF valuation
  5. AI thesis and insights
  6. Assemble a fully sourced report
Deployed with Coolify on Hetzner
React app
  • Live progress while it generates
  • Tabbed report viewer
  • Accounts, guest reports and share links
  • Admin dashboard
Postgres · Supabase
  • Reports and profiles
  • Row-level security
  • Two-tier cache

Four decisions behind the trail.

  1. 01

    Every value carries its source

    Each number is a sourced value: a link to the filing, the market-data source, or the formula and inputs behind a calculation.

  2. 02

    Filings change their tags

    Companies rename XBRL concepts between years. Extraction checks every candidate tag and keeps the one with the most recent data.

  3. 03

    Access rules live in the database

    Row-level security keeps reports private to their owner, lets share links read one report, and reads roles fresh on every request.

  4. 04

    Repeat tickers are nearly free

    A shared cache with per-source expiry brings a repeat report down from about 13 market-data calls to between zero and two.

See the startup journey

Product interfaces

Amala Vista

Pick the apartment on the building itself.

Buyers explore the units on the building itself, filtering by rooms, floor, size and availability.

polygon-showcase.vercel.app↗
Amala Vista website screenshot
The problem
Developers often pre-sell apartments before a building is finished, with little more than floor plans.
What I did
Owned the full cycle: the idea and the problem, the solution, building it in React, Next.js and Tailwind, deploying it, and the enrichment and outreach behind it.
The outcome
Outreach across the Balkans, Turkey and Spain found some interest, but no company committed to moving forward.
Explore Amala Vista

Data pipelines

1,500+companies processed

From a company list
to a ranked shortlist.

Built for Amala Vista’s outreach across the Balkans and Turkey. Companies were enriched and ranked so outreach started with the ones most likely to need the product.

  1. 01 / Input

    Company lists

    Raw lists of companies across the region.

  2. 02 / Enrich

    Extruct API

    Fill in the company data the lists were missing.

  3. 03 / Research

    Claude Code CLI

    Scrape and read each company’s website.

  4. 04 / Score

    ICP scoring in Python

    Rank every company against the ideal customer profile.