Feature/issue 34 pipeline#243
Open
Subhra-Nandi wants to merge 12 commits into
Open
Conversation
- Add PerceptionBus with backpressure (evict oldest frame when full) - Add ContextEngine, IntelligenceCore, TrustScorer stubs - Add GuidanceDispatcher with deduplication (SHA-1 fingerprint) - Add ActionLogger for undo/replay audit trail - Add PipelineMetrics with p50/p95/p99 latency tracking - Target: ≤500 ms frame-to-WebSocket latency
…int (sahoo-tech#34) - Add asynccontextmanager lifespan that starts/stops ExecraPipeline - Add GET /health and GET /metrics REST endpoints - Add POST /frame for synthetic frame injection (testing/demos) - Add WebSocket /ws/guidance for real-time guidance broadcast - Pipeline injected from main.py in production; self-hosted in standalone mode
…sahoo-tech#34) - Parse --domain, --mode, --host, --port, --log-level CLI args - Inject ExecraPipeline into api.main before uvicorn starts - Register SIGINT/SIGTERM handlers for clean shutdown - asyncio.gather(server.serve(), pipeline.run()) shares the event loop
Component tests: - PerceptionBus: put/get frame, backpressure eviction - GuidanceDispatcher: send to client, deduplication, multi-instruction - TrustScorer: high/low confidence threshold Integration tests (full pipeline, mocked LLM): - Single frame produces guidance on WS client - All frames processed within 500 ms SLA - Backpressure: 50-frame flood never blocks - Deduplication: 5 identical frames → 1 WS message - None guidance: no WS message sent - Clean stop is idempotent - Metrics populated correctly after run
…tech#34) Added Optional from yping to core/config.py. The type annotation REDIS_AUTH: Optional[str] = None on line 54 was raising NameError: name 'Optional' is not defined at import time, which blocked pytest from loading conftest.py and prevented all tests from running.
- benchmark_pipeline.py: measures p50/p95/p99 latency under variable worker counts and frame injection rates - docs/benchmark_results.md: stub-LLM results (p99 < 2 ms), production estimates with real LLM (~150-460 ms), SLA compliance statement and strategies (streaming, prompt caching, parallel rules)
…features (sahoo-tech#150) Safety bug fix: - Replace if/elif/elif chain with independent if blocks so ALL hazards in a single frame are reported; old code silently dropped any rule after the first matching elif (e.g. power_tool warning invisible when circuit_board also present) New features: - Implement _check_soldering_proximity(): Euclidean bbox-centre distance; fires specific component name when iron within 150px threshold; graceful fallback when no bbox data provided - Implement _select_template(): scores each template by keyword matches against detected_classes; returns best match in active_template field - Add active_template field to GuidanceInstruction dataclass - Implement _parse_ocr(): extracts repair step numbers and P/N part numbers from OCR text; appended as WARNING-level guidance notes - Document hand_results parameter (reserved for future gesture rules) Tests: 4 → 24 - Keep all 4 original tests; all still pass unchanged - Add TestCircuitBoardRule, TestPowerToolRule, TestSolderingIronRule positive and negative path tests - Add TestMultipleSimultaneousHazards: regression tests for the elif bug - Add TestTemplateSelection: all 4 templates + no-match case - Add TestOCRIntegration: step number, part number, empty string - Add TestComponentIdentification: mapping, unknown, deduplication - Add TestEdgeCases: empty list, uppercase class, missing key, None Docs: - Expand docs/physical_domain.md: component dictionary table, step templates table, safety rules with trigger conditions and exact messages, status codes, OCR integration, GuidanceInstruction fields
…missing features (sahoo-tech#150)" This reverts commit f28eab5.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🔗 Related Issue
Closes #34
📝 Summary of Changes
Implements the complete end-to-end real-time guidance streaming pipeline
as specified in issue #34. Connects every module into a single async loop:
PerceptionBus → ContextEngine → IntelligenceCore → TrustScorer → GuidanceDispatcher → WebSocket
Also fixes the original submission which replaced existing
api/main.pyandmain.pycontent entirely instead of integrating the pipeline cleanly alongsidethe existing code.
🔍 Type of Change
🧪 How Was This Tested?
All 14 integration tests pass. Tests cover:
Test environment:
📸 Screenshots / Recordings (if applicable)
Benchmark results (stub LLM, Windows):
SLA target: ≤500 ms. Status: ✅ compliant
✅ Pre-Submission Checklist
upstream/maindocs/benchmark_results.md).envfiles, secrets, or model weights💬 Additional Notes for Reviewer
What changed since the previous submission (closed PR):
api/main.py— previous version replaced the entire file, deleting allexisting routes (
status,mode,actions,context), CORS config usingsettings.CORS_ORIGINS, and the root/endpoint. Fixed by keeping allexisting content and adding pipeline startup via
asynccontextmanager lifespan.main.py— previous version replaced the entire file, deleting the banner,settings, and logging setup. Fixed by keeping all original content and adding
async def _run()withasyncio.gather(server.serve(), pipeline.run()).An unrelated commit for issue Build a Hardware Repair Domain Guidance Engine #150 (hardware engine) accidentally landed on
this branch — it has been cleanly reverted.
Design decisions:
_ingest_loopevicts oldest frame when processing queueis full; perception is never awaited on a full queue.
never dispatched twice in a row.
time.perf_counter()used overtime.monotonic()fornanosecond resolution on Windows. p50/p95/p99 exposed via
GET /metrics.stop()cancels all tasks, tears down subsystems inreverse-start order, idempotent (safe to call twice).
ContextEngine._infer_task_type,IntelligenceCore._build_instruction, andPerceptionBus._capture_loopare clearly marked stubs ready to be swapped with real drivers.