Release Notes - v1.0.2

This release is a deep pass on the bulk CV pipeline — both the reliability of turning a folder of CVs into candidates, and the visibility into what happened to every file. CV text extraction now runs on Mistral OCR, the resume classifier no longer drops valid CVs when the model's JSON is truncated, and large CVs and late-arriving email attachments process reliably thanks to download retries and orphaned-file cleanup. The Upload History dialog got a full reconciliation and UX overhaul: every submitted file is now accounted for — processed, rejected, failed, still working, or never received — with counts that always sum to the submitted total, plus a compact batch selector, a clear file filter, an expand-to-near-full-screen toggle, and a "View Upload History" jump straight from the Activity panel. Stuck CV-processing jobs can now be cancelled and retried instead of spinning forever. On the assistant side, chat list links are built server-side, the AI-SQL safety-keyword check matches on word boundaries, and deep CV analysis is scoped to the company.

• Mistral OCR for CV extraction — CV text extraction moves to Mistral OCR, replacing the previous Tesseract + vision pipeline for more accurate, consistent parsing (including Arabic and mixed-language CVs)
• Upload History reconciliation — every file you submit is now accounted for in one place; counts for processed / rejected / failed / in-progress / not-received always add up to the submitted total, and files that never reached the server are surfaced explicitly
• Upload History UX refresh — a compact batch selector, a clear file filter with per-state counts, an expand/collapse (near-full-screen) toggle, and a "View Upload History" link on bulk CV upload items in the Activity panel
• CV processing recovery — stuck CV-processing jobs can be cancelled mid-run and retried, so an interrupted batch no longer stays stuck

• Idle Upload History is quiet — the dialog refreshes live only while an upload is actively processing and stops once every file reaches a terminal state, resuming when a new upload starts
• Clearer failure reasons — each failed file shows an actionable reason (e.g. "No email address found in this CV")

• CV classifier no longer drops valid CVs — a truncated JSON response from the resume-header classifier no longer causes valid CVs to be silently discarded
• Large CV & late-content email handling — large CVs and emails whose content arrives late now process reliably, with a download retry and cleanup of orphaned S3 objects
• Chat AI assistant — candidate/job list links are built server-side (no more corrupted links), the AI-SQL safety-keyword check matches on word boundaries (fewer false blocks), and deep CV analysis is scoped to the caller's company

More accurate text extraction across CV formats

CV text extraction now runs on Mistral OCR, replacing the previous Tesseract + vision approach. The result is more accurate and consistent parsing across scanned PDFs, exported documents, and mixed English/Arabic CVs — which feeds cleaner data into candidate creation, AI review, and search.

In testing, a batch of 26 real CVs (including scanned and Arabic documents, and PDFs up to ~2 MB) extracted full candidate details — name, email, phone, and address — with the correct values pulled even from right-to-left layouts.

Every submitted file is accounted for

The Upload History dialog now reconciles the full lifecycle of a batch. For any upload you can see exactly where every file ended up, and the numbers always add up to what you submitted.

The counts are derived from the batch's authoritative submitted total, so they always sum correctly — a file can't silently "disappear."

A completed batch: 26 submitted = 17 processed + 9 failed, with a proportional bar and a filter that shows per-state counts

Filtering to Failed shows each file with an actionable reason — here, "No email address found in this CV"

Easier to navigate, quieter when idle

The dialog and its Activity-panel entry point got a usability pass.

While a batch runs: the batch selector, live per-stage sections (Processing / Queued / Processed / Failed), a running completed/failed/remaining tally, and a Cancel control

The expand toggle opens the dialog to near-full-screen so large batches are easier to scan; a second click restores it

An upload item in the Activity panel expands to a per-file summary with a View Upload History link straight to its batch

Cancel a stuck batch, retry the rest

If a CV-processing job gets stuck or you want to stop a run partway, you can now cancel it mid-process — files already processed keep their candidates, and the remaining files are skipped. Interrupted or failed processing can be retried so a batch no longer stays stuck with no way forward.

The Cancel control is available both on the bulk-upload activity item and inside the Upload History dialog while a batch is processing.