September 23, 2026 / Ashwin Sharma
Eternal Web built Oratora, a multi-tenant Call for Papers platform, for R G Manek and Associates, a Chartered Accountancy firm that organizes professional and business conferences, including Tech Expo Gujarat, for its community. Oratora is live in production at oratora.in and uses Anthropic’s Claude to pre-screen conference talk submissions before human reviewers see them.
Conference program committees typically work with a small volunteer review team under tight deadlines. Reviewers need to separate submissions that genuinely match the event’s announced topics and quality bar from weaker ones. That’s an objective, rules-based first pass, well suited to AI-assisted screening, done before reviewers spend their limited time on deeper judgment calls.
A multi-tenant CFP platform managing the full talk lifecycle, from speaker submission through schedule publication, with four dashboards: Speakers, Organizers, Agencies (third-party speaker bureaus and partners who submit programmatically via API), and Admins.
claude-haiku-4-5-20251001) via @anthropic-ai/sdkWhen a submission arrives, the organizer triggers an AI evaluation with one click. The prompt runs in two explicit phases: first, a topic-alignment check against the CFP’s accepted topics (off-topic talks are capped at a low score regardless of writing quality); second, only for on-topic submissions, a quality pass scoring abstract clarity, originality, speaker credibility, and format fit. Claude returns a 1–10 score with a short plain-English summary.
Prompt instructions alone can be ignored by a model, so Oratora doesn’t rely on them. After Claude responds, the application independently re-applies the same topic-match caps in code. If the model ever returns a high score for an off-topic submission, the server corrects it before it’s stored. The same layer includes fault-tolerant JSON parsing (markdown-fence stripping, regex fallback) and graceful degradation: if the Claude API key is ever unavailable, the scoring endpoint returns a clear error while the rest of the platform keeps working normally.
AI scores are stored in their own database fields, completely separate from manual star ratings and reviewer accept/reject decisions. They are never averaged together. Scoring is one-time per submission by design, so the AI result stays a stable reference point through the review process.
Third-party speaker bureaus and partner organizations can submit on behalf of their speakers through a REST API, authenticated with bcrypt-hashed API keys (12-character prefix lookup) and per-hour rate limiting. SSO login is available via short-lived (5-minute) signed JWT redirects. Every API call is logged for auditability, and webhook deliveries (fired on submission created, updated, and withdrawn events) are signed with HMAC-SHA256 for verifiable, tamper-evident notifications, with automatic retries on failed delivery.
Oratora completed its first live CFP round for Tech Expo Gujarat, processing 13 talk submissions. Every submission was pre-screened by Claude before human review, with AI scores ranging 3–7 out of 10, giving the organizer an immediate signal alongside independent manual star ratings from the review team.
The two scores don’t always agree, and that’s by design. One submission scored 3/10 from Claude’s pre-screening, yet the human reviewer rated it 5/5 and accepted it after reading the full abstract. That’s the system working as intended: Claude surfaces a fast first read, but the organizer’s own judgment decides the outcome. The AI didn’t need to be right, only useful as a starting point.
Manually reviewing a CFP submission (reading the abstract, checking speaker credentials, and applying the scoring rubric) typically takes a reviewer about 5 minutes. Oratora’s AI pre-screening returns a 0–10 relevance score in about 2 seconds, roughly 150x faster on the first pass. Across the first live round’s 13 submissions, this cut initial screening effort from roughly 65 minutes of manual work down to under 30 seconds of AI processing, freeing reviewers to spend their time on deeper evaluation of borderline and high-potential talks, like the submission Claude scored 3/10 that a human reviewer overrode to 5/5 and accepted.
Oratora’s AI scoring dashboard from the first live CFP round for Tech Expo Gujarat. Each submission shows its Claude-generated score (1–10) alongside the human reviewer’s star rating and final status.
“Oratora has made our CFP review process significantly faster and more consistent. The AI pre-screening layer helps our reviewers focus on genuinely relevant submissions instead of manually filtering out off-topic ones.”
— Rushabh Manek, Director, R G Manek and Associates
AWS Certified Solution Architect with a proven track record in shaping information management strategies. Recognized for expertise in designing and implementing scalable, highly available, and fault-tolerant solutions aligned with enterprise objectives. Accomplished Solutions Architect with a demonstrated history in information technology. Proficient in Amazon Web Services (AWS), WHM, and cPanel.
Have queries about your project idea or concept? Please drop in your project details to discuss with our AWS Global Cloud Infrastructure service specialists and consultants.
