Skip to content
Yetmgeta Redahegn
All work

Built for sequa

Agentic Funding-Application Assistant

Turns a small-business owner's spoken story, phone photos and paper licence into a complete, honest funding application, and helps reviewers produce a ranked shortlist they can defend.

Role
AI Engineer
Status
Status: Delivered
Stack
  • LangChain
  • FastAPI
  • Next.js
  • LLMs
  • Multilingual

The problem

Funding for small Ethiopian businesses exists, but the application doesn't fit the people it's meant for. It asks eighteen sub-questions: five years of sales and employment split by gender and age, a management table, an organogram, a machinery list and fifteen declarations. A 100-point grid then scores it on nine weighted criteria, with three exclusion factors that end an application on the spot.

A workshop owner with a feature phone can't do that alone, so she stays out. On the other side, a reviewer scores each batch by hand.

What I built

  • Applicant path

    An intake agent that sits between someone who talks and a system that needs structured records. It takes a spoken story plus photos of the business licence and the workshop, and fills in the application form and a project draft (title, location, SDGs, funding target in ETB, beneficiaries, milestones, sector).

  • Honest by design

    Every field that isn't established goes on a gap list (what's missing, what it needs, and from whom) instead of being guessed.

  • Declarations

    Explained to the applicant in her own language, with a record that she understood. The agent never ticks a declaration on her behalf.

  • Eligibility and scoring

    Runs the eligibility gate and the weighted grid, gives reasoning for each criterion, and names any exclusion factors.

  • Reviewer path

    Takes a batch of applications, routes each to the right grid variant, scores it, and returns a ranked shortlist. Each company gets a justification paragraph, open questions for a site visit, and a list of self-contradictions (for example, a licence date that conflicts with the years in operation, or ownership percentages that don't add up).

  • Multilingual

    Works in English, Amharic, German and other languages.

Architecture

  1. Voice note + photos
  2. Intake agent (LangChain)
  3. Structured application + project draft
  4. Eligibility gate & weighted grid (AI evaluation / decision step)
  5. Gap list & declarations
  6. Reviewer agent (AI evaluation / decision step)
  7. Ranked shortlist

Green marks an AI evaluation / decision step.

Engineering highlights

  • Agentic design with separate applicant and reviewer paths over one shared application schema
  • Unverified fields are flagged, never guessed
  • Per-criterion reasoning, so every score can be explained and defended

Contact

Have something to build?

Tell me what you want automated or which AI feature you need, and I'll reply with a concrete plan.

Addis Ababa, Ethiopia (UTC+3). Full overlap with Europe, morning overlap with the US.