AI Resume Builder
LLM-crafted resumes and cover letters from a single structured profile.
1 → N
one profile generates tailored resumes per job description
Parsed
intelligent import of existing CV content
Automated
matching cover letters alongside each resume
Tech Stack
The Challenge
Generic resume templates lose to tailored ones in every screen. The hard part was not rendering a PDF — it was getting an LLM to rewrite a candidate's real experience for a specific job description without inventing facts, and keeping the output editable.
Approach
- 1
Structured the profile as discrete, schema-validated records (experience, skills, projects) so the generation step rewrites facts it cannot fabricate.
- 2
Used LangChain prompt chains: parse → match against the job description → draft per section → assemble, with each stage independently retryable.
- 3
Stored drafts and versions in MongoDB so users could iterate on generations without losing prior work.
- 4
Kept the final document editable — AI output is a draft, and the UI treats it that way.
Technical Highlights
Grounded generation
Prompts were constrained to the user's own records, with a verification pass flagging claims that had no source in the profile — the fastest way to kill hallucinated skills.
Section-wise chains
Generating summary, experience bullets, and cover letter as separate chain steps kept outputs consistent and made regeneration cheap (only the rejected section re-runs).
What I Learned
LLM tools succeed when the data model constrains them; free-form prompting produces free-form nonsense.
Versioned drafts turn AI generation from a gamble into an iteration loop users trust.