:: Browse bounties
Open bounties
14 open · escrow-secured · claim one and ship it.
RAG pipeline over Notion + Slack
“We want an internal Q&A over our Notion workspace + last 90 days of Slack. Chunk + embed (we have pgvector), retrieve top-k with citations, stream answers via SSE. Handle access control loosely (one shared index is fine). Deliver a FastAPI service + a tiny eval set (~40 Q/A).”
Fine-tune Llama 3.1 for support triage
“Classify inbound support tickets into ~40 product queues. We have 60k labeled tickets. Fine-tune a Llama-3.1-8B LoRA, ship an eval harness, hit >92% top-1 accuracy on our held-out set. Deliver adapter weights + inference script.”
Migrate Next.js 14 → 16 App Router
“Move ~12 pages-router routes to the App Router, fix the caching/ISR behavior on 3 dynamic routes, and keep our deploy zero-downtime. Next 16 + Turbopack. Tests already green; don't regress them.”
Stripe webhook idempotency audit
“Audit our Stripe webhook handler for replay/double-capture bugs. Add event-id idempotency, signature verification, a reconciliation cron, and a status-guarded capture path. Write up the threats you closed.”
Cursor rules pack for a Turborepo
“Write a set of .cursor/rules for our TS Turborepo: how to run tests per package, our commit convention, safe-refactor patterns, and when to touch shared packages. Should make new hires productive in Cursor fast.”
Agent that books flights from natural language
“A tool-using agent that takes 'flights to Tokyo mid-March under $1k' and returns bookable options via our flights API. Handle date ambiguity, multi-city, and confirm-before-booking. Ship a working CLI demo + a short loom.”
Whisper batch transcoder + diarization
“Batch transcribe podcast episodes with Whisper-large-v3 + speaker diarization. GPU-efficient (we have 1 A10G), S3 in / S3 out, emit timestamped JSON. Throughput target: 20 hrs of audio / hr.”
Linear ↔ GitHub two-way sync
“Two-way sync between Linear and GitHub issues for 3 repos. Handle status mapping, label conflicts, and comment threading. Ship as a Vercel function (webhook) + a reconciliation cron.”
Eval harness for a RAG retrieval step
“Build an eval harness that scores our RAG retrieval (recall@k, MRR, faithfulness) across chunk-size and embedding configs. Output a ranked report so we can pick a default. ~80 golden Q/A.”
Slack bot that summarizes threads
“A Slack slash-command that summarizes a thread in 5 bullets, lists decisions, and @-mentions owners. Bolt-on to our existing Slack app. Keep token cost reasonable.”
Quantize a 70B model for a single 4090
“Quantize our 70B base to run on a single 24GB 4090. Target >40 tok/s at acceptable quality (perplexity within +5%). Ship an inference server (vLLM/llama.cpp) + the quant recipe.”
Tailwind v4 migration for a design system
“Migrate our 30-component React design system from Tailwind v3 to v4 (CSS-first config). Preserve our design tokens and component APIs. No visual regressions (we have Playwright snapshots).”
PII redaction layer for support transcripts
“Add a redaction layer that scrubs emails, phone numbers, and card numbers from support transcripts before they reach our data lake. Keep a reversible map in a sealed store. Precision target: zero leaks on our 1k-sample eval.”
Dockerize + CI for a FastAPI ML service
“Production-grade multi-stage Dockerfile for our FastAPI ML service (CUDA base, slim runtime) + a GitHub Actions workflow that builds, tests, and pushes on every PR. Keep image < 2.5GB.”